In [21]:
show_images(train_loader)
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
import os, sys, copy, helper, warnings
warnings.filterwarnings('ignore')
from tqdm.notebook import tqdm
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
%matplotlib inline
from PIL import Image
import cv2
import csv
import torch
from torch import nn
import torch.nn.functional as F
from torch.utils.data import Dataset, DataLoader
from torch.utils.data.sampler import SubsetRandomSampler
import torch.optim as optim
import torchvision
import torchvision.transforms as transforms
from torchvision import datasets, models, utils
from torchvision.utils import make_grid
import imageio
from sklearn.metrics import confusion_matrix
from skimage.io import imread, imshow, imread_collection, concatenate_images
from skimage.transform import resize
from skimage.morphology import label
def show_images(dataloader):
for images, labels in dataloader:
fig, ax = plt.subplots(figsize=(16,16))
ax.set_xticks([]); ax.set_yticks([])
ax.imshow(make_grid(images, nrow=5).permute(1, 2, 0))
break
def accuracy(output, target):
output = torch.exp(output)
top_p,top_class = output.topk(1,dim = 1)
equals = top_class == target.view(*top_class.shape)
return torch.mean(equals.type(torch.FloatTensor))
train_on_gpu = torch.cuda.is_available()
train_on_gpu
True
!nvidia-smi
/bin/bash: /home/sysadm/anaconda3/lib/libtinfo.so.6: no version information available (required by /bin/bash)
Sat Jun 25 03:19:39 2022
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 515.48.07 Driver Version: 515.48.07 CUDA Version: 11.7 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|===============================+======================+======================|
| 0 Quadro GV100 On | 00000000:2D:00.0 On | Off |
| 39% 52C P0 47W / 250W | 309MiB / 32768MiB | 0% Default |
| | | N/A |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=============================================================================|
| 0 N/A N/A 1425 G /usr/lib/xorg/Xorg 39MiB |
| 0 N/A N/A 2117 G /usr/lib/xorg/Xorg 135MiB |
| 0 N/A N/A 2232 G ...mviewer/tv_bin/TeamViewer 5MiB |
| 0 N/A N/A 2256 G /usr/bin/gnome-shell 43MiB |
| 0 N/A N/A 2862 G ...034559963964204169,131072 65MiB |
+-----------------------------------------------------------------------------+
path = '/home/sysadm/Documents/3ml_kkm/env/chest_xray'
train_dir = path + '/train'
test_dir = path + '/test'
n_train = []
n_test = []
classes = ['PNEUMONIA', 'NORMAL']
for c in classes:
train_images = os.listdir(train_dir + '/' + c)
n_train.append(len(train_images))
test_images = os.listdir(test_dir + '/' + c)
n_test.append(len(test_images))
print(n_train)
print(n_test)
[3884, 1349] [390, 235]
data = pd.DataFrame({'Classes' : classes,
'Train' : n_train,
'Test' : n_test})
print(data)
Classes Train Test 0 PNEUMONIA 3884 390 1 NORMAL 1349 235
# plt.figure(figsize=(8,8))
# sns.barplot(data=data, x='Classes', y='Train')
# plt.xlabel('Classes',fontsize=15)
# plt.ylabel('Count',fontsize=15)
# plt.title('Training data',fontsize=15);
# plt.figure(figsize=(8,8))
# sns.barplot(data=data, x='Classes', y='Test')
# plt.xlabel('Classes',fontsize=15)
# plt.ylabel('Count',fontsize=15)
# plt.title('Test data',fontsize=15);
# train set
ptrain = []
ctrain = []
for folder_name in classes:
for files in os.listdir(train_dir + '/' + folder_name):
if files.split(".")[-1].lower() in {"jpeg", "jpg", "png"}:
path = train_dir + '/' + folder_name + '/' + files
clss = folder_name
ptrain.append(path)
ctrain.append(clss)
# test set
ptest = []
ctest = []
for folder_name in classes:
for files in os.listdir(test_dir + '/' + folder_name):
if files.split(".")[-1].lower() in {"jpeg", "jpg", "png"}:
path = test_dir + '/' + folder_name + '/' + files
clss = folder_name
ptest.append(path)
ctest.append(clss)
train_df = pd.DataFrame({'Path':ptrain,'Class':ctrain})
test_df = pd.DataFrame({'Path':ptest,'Class':ctest})
train_df.head(10)
| Path | Class | |
|---|---|---|
| 0 | /home/sysadm/Documents/3ml_kkm/env/chest_xray/... | PNEUMONIA |
| 1 | /home/sysadm/Documents/3ml_kkm/env/chest_xray/... | PNEUMONIA |
| 2 | /home/sysadm/Documents/3ml_kkm/env/chest_xray/... | PNEUMONIA |
| 3 | /home/sysadm/Documents/3ml_kkm/env/chest_xray/... | PNEUMONIA |
| 4 | /home/sysadm/Documents/3ml_kkm/env/chest_xray/... | PNEUMONIA |
| 5 | /home/sysadm/Documents/3ml_kkm/env/chest_xray/... | PNEUMONIA |
| 6 | /home/sysadm/Documents/3ml_kkm/env/chest_xray/... | PNEUMONIA |
| 7 | /home/sysadm/Documents/3ml_kkm/env/chest_xray/... | PNEUMONIA |
| 8 | /home/sysadm/Documents/3ml_kkm/env/chest_xray/... | PNEUMONIA |
| 9 | /home/sysadm/Documents/3ml_kkm/env/chest_xray/... | PNEUMONIA |
train_df.to_csv('train_data.csv',index=False)
test_df.to_csv('test_data.csv',index=False)
train_transforms = transforms.Compose([transforms.RandomRotation(30),
transforms.Resize((256,256)),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize((0.5,0.5,0.5),(0.5,0.5,0.5))
])
test_transforms = transforms.Compose([transforms.Resize((256,256)),
transforms.ToTensor(),
transforms.Normalize((0.5,0.5,0.5),(0.5,0.5,0.5))
])
class PneumoniaDataset():
def __init__(self,df,labels,transform=True):
self.df = df
self.labels=list(df.iloc[:,1])
self.transform = transform
def __len__(self):
return len(self.df)
def __getitem__(self,idx):
row = self.df.loc[idx]
img_file, img_class = row['Path'], row['Class']
img = Image.open(img_file).convert('RGB')
if self.transform:
img = self.transform(img)
return img,img_class
class_dict = {'NORMAL':0, 'PNEUMONIA':1}
train_data = PneumoniaDataset(train_df,class_dict,transform = train_transforms)
test_data = PneumoniaDataset(test_df,class_dict,transform = test_transforms)
# verifying the dataloader
print(len(train_data))
next(iter(train_data))
5232
(tensor([[[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
...,
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.]],
[[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
...,
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.]],
[[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
...,
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.]]]),
'PNEUMONIA')
print(len(test_data))
next(iter(test_data))
624
(tensor([[[-0.6549, -0.6941, -0.6941, ..., -0.6157, -0.6314, -0.6471],
[-0.6863, -0.7020, -0.6941, ..., -0.6235, -0.6314, -0.6471],
[-0.6863, -0.6863, -0.6941, ..., -0.6235, -0.6314, -0.6471],
...,
[-0.8196, -0.8275, -0.8275, ..., -0.8353, -0.8275, -0.8196],
[-0.3333, -0.8039, -0.8275, ..., -0.8353, -0.8039, -0.3255],
[-0.8510, -0.8275, -0.8275, ..., -0.8353, -0.8353, -0.8510]],
[[-0.6549, -0.6941, -0.6941, ..., -0.6157, -0.6314, -0.6471],
[-0.6863, -0.7020, -0.6941, ..., -0.6235, -0.6314, -0.6471],
[-0.6863, -0.6863, -0.6941, ..., -0.6235, -0.6314, -0.6471],
...,
[-0.8196, -0.8275, -0.8275, ..., -0.8353, -0.8275, -0.8196],
[-0.3333, -0.8039, -0.8275, ..., -0.8353, -0.8039, -0.3255],
[-0.8510, -0.8275, -0.8275, ..., -0.8353, -0.8353, -0.8510]],
[[-0.6549, -0.6941, -0.6941, ..., -0.6157, -0.6314, -0.6471],
[-0.6863, -0.7020, -0.6941, ..., -0.6235, -0.6314, -0.6471],
[-0.6863, -0.6863, -0.6941, ..., -0.6235, -0.6314, -0.6471],
...,
[-0.8196, -0.8275, -0.8275, ..., -0.8353, -0.8275, -0.8196],
[-0.3333, -0.8039, -0.8275, ..., -0.8353, -0.8039, -0.3255],
[-0.8510, -0.8275, -0.8275, ..., -0.8353, -0.8353, -0.8510]]]),
'PNEUMONIA')
# training indices to be used for validation
num_train = len(train_data); print(num_train)
indices = list(range(num_train))
np.random.shuffle(indices)
split = int(np.floor(0.2 * num_train)); print(split)
train_idx, valid_idx = indices[split:], indices[:split]
5232 1046
# defining samplers for obtaining training and validation batches
train_sampler = SubsetRandomSampler(train_idx)
valid_sampler = SubsetRandomSampler(valid_idx)
# loading the data
train_loader = torch.utils.data.DataLoader(train_data, batch_size = 30, sampler = train_sampler)
valid_loader = torch.utils.data.DataLoader(train_data, batch_size = 30, sampler = valid_sampler)
test_loader = torch.utils.data.DataLoader(test_data, batch_size = 30, shuffle = True)
next(iter(train_loader))
[tensor([[[[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
...,
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.]],
[[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
...,
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.]],
[[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
...,
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.]]],
[[[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
...,
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.]],
[[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
...,
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.]],
[[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
...,
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.]]],
[[[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
...,
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.]],
[[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
...,
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.]],
[[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
...,
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.]]],
...,
[[[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
...,
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.]],
[[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
...,
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.]],
[[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
...,
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.]]],
[[[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
...,
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.]],
[[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
...,
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.]],
[[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
...,
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.]]],
[[[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
...,
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.]],
[[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
...,
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.]],
[[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
...,
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.],
[-1., -1., -1., ..., -1., -1., -1.]]]]),
('PNEUMONIA',
'PNEUMONIA',
'PNEUMONIA',
'PNEUMONIA',
'PNEUMONIA',
'PNEUMONIA',
'NORMAL',
'PNEUMONIA',
'PNEUMONIA',
'PNEUMONIA',
'PNEUMONIA',
'PNEUMONIA',
'PNEUMONIA',
'PNEUMONIA',
'PNEUMONIA',
'PNEUMONIA',
'NORMAL',
'PNEUMONIA',
'PNEUMONIA',
'NORMAL',
'PNEUMONIA',
'NORMAL',
'NORMAL',
'NORMAL',
'PNEUMONIA',
'PNEUMONIA',
'NORMAL',
'PNEUMONIA',
'PNEUMONIA',
'PNEUMONIA')]
show_images(train_loader)
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
show_images(valid_loader)
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
show_images(test_loader)
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
class ConvNet(nn.Module):
def __init__(self):
super().__init__()
self.conv1 = nn.Conv2d(3, 16, 3, padding = 1)
self.conv2 = nn.Conv2d(16, 32, 3, padding = 1)
self.conv3 = nn.Conv2d(32, 64, 3, padding = 1)
self.conv4 = nn.Conv2d(64, 128, 3, padding = 1)
self.conv5 = nn.Conv2d(128, 256, 3, padding = 1)
self.bn1 = nn.BatchNorm2d(16)
self.bn2 = nn.BatchNorm2d(32)
self.bn3 = nn.BatchNorm2d(64)
self.bn4 = nn.BatchNorm2d(128)
self.bn5 = nn.BatchNorm2d(256)
self.pool = nn.MaxPool2d(2, 2)
self.fc1 = nn.Linear(256 * 8 * 8, 1024)
self.fc2 = nn.Linear(1024, 512)
self.fc3 = nn.Linear(512, 2)
self.dropout = nn.Dropout(0.6)
def forward(self, x):
x = self.pool(F.relu(self.bn1(self.conv1(x))))
x = self.pool(F.relu(self.bn2(self.conv2(x))))
x = self.pool(F.relu(self.bn3(self.conv3(x))))
x = self.pool(F.relu(self.bn4(self.conv4(x))))
x = self.pool(F.relu(self.bn5(self.conv5(x))))
x = x.view(-1, 256 * 8 * 8)
x = self.dropout(x)
x = F.relu(self.fc1(x))
x = self.dropout(x)
x = F.relu(self.fc2(x))
x = self.dropout(x)
x = F.log_softmax(self.fc3(x), dim=1)
return x
We have trained this model twice - with 2 different optimisers:
These are known to produce the best results in image classification problems. The training given below has been done on Adam. To use SGD with Nesterov accelaration instead, use the commented out optimizer instead of the one used below.
# specifying the loss function (categorical cross-entropy)
model = ConvNet()
criterion = nn.CrossEntropyLoss()
# specifying optimizer
optimizer = optim.Adam(model.parameters(), lr = 0.008, weight_decay = 0.02)
# optimizer = optim.SGD(model.parameters(), lr = 0.008,momentum = 0.9, nesterov = True)
# specifying scheduler
scheduler = optim.lr_scheduler.ExponentialLR(optimizer,gamma = 0.2)
if train_on_gpu:
model.cuda()
print(model)
ConvNet( (conv1): Conv2d(3, 16, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)) (conv2): Conv2d(16, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)) (conv3): Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)) (conv4): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)) (conv5): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)) (bn1): BatchNorm2d(16, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (bn2): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (bn3): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (bn4): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (bn5): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True) (pool): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False) (fc1): Linear(in_features=16384, out_features=1024, bias=True) (fc2): Linear(in_features=1024, out_features=512, bias=True) (fc3): Linear(in_features=512, out_features=2, bias=True) (dropout): Dropout(p=0.6, inplace=False) )
pwd
'/home/sysadm/Documents/3ml_kkm/env'
# number of epochs to train the model
n_epochs = 100
valid_loss_min = np.Inf # track change in validation loss
train_losses, valid_losses = [], []
train_acc, valid_acc = [], []
for epoch in tqdm(range(1, n_epochs+1), desc = "Train Model/Epoch: "):
# keeping track of loss and accuracy
train_loss = 0.0
valid_loss = 0.0
train_accuracy = 0.0
valid_accuracy = 0.0
######################
# training the model #
######################
model.train()
for data, target in tqdm(train_loader, desc = 'Model Train: '):
# moving tensors to GPU if CUDA is available
data = torch.Tensor(data)
target = tuple([0 if i=='NORMAL' else 1 for i in list(target)])
target = torch.Tensor(np.array(target))
if train_on_gpu:
data, target = data.cuda(), target.cuda()
# clearing the gradients of all optimized variables
optimizer.zero_grad()
# forward pass: compute predicted outputs by passing inputs to the model
output = model(data)
target = target.long()
# calculating the batch loss
loss = criterion(output, target)
# backward pass: compute gradient of the loss with respect to model parameters
loss.backward()
# performing a single optimization step (parameter update)
optimizer.step()
# updating training loss
train_loss += loss.item()*data.size(0)
# updating training accuracy
train_accuracy += accuracy(output,target)
scheduler.step() # decay LR
########################
# validating the model #
########################
model.eval()
for data, target in tqdm(valid_loader, desc = 'Model validation: '):
# moving tensors to GPU if CUDA is available
data = torch.Tensor(data)
target = tuple([0 if i=='NORMAL' else 1 for i in list(target)])
target = torch.Tensor(np.array(target))
if train_on_gpu:
data, target = data.cuda(), target.cuda()
# forward pass: compute predicted outputs by passing inputs to the model
output = model(data)
target = target.long()
# calculating the batch loss
loss = criterion(output, target)
# updating average validation loss
valid_loss += loss.item()*data.size(0)
# updating validation accuracy
valid_accuracy += accuracy(output,target)
# calculating average losses
train_loss = train_loss/len(train_loader)
valid_loss = valid_loss/len(valid_loader)
# calculating average accuracy
train_accuracy = train_accuracy/len(train_loader)
valid_accuracy = valid_accuracy/len(valid_loader)
train_losses.append(train_loss)
valid_losses.append(valid_loss)
train_acc.append(train_accuracy)
valid_acc.append(valid_accuracy)
# printing training/validation statistics
print('Epoch: {} \tTraining Loss: {:.6f} \tValidation Loss: {:.6f} \n\t\tTraining Accuracy: {:.6f} \tValidation Accuracy: {:.6f}'.format(
epoch, train_loss, valid_loss, train_accuracy, valid_accuracy))
# saving model if validation loss has decreased
if valid_loss <= valid_loss_min:
print('Validation loss decreased ({:.6f} --> {:.6f}). Saving model ...'.format(
valid_loss_min,
valid_loss))
torch.save(model.state_dict(), './../models/model_pneumonia.pt')
valid_loss_min = valid_loss
Epoch: 1 Training Loss: 78.118760 Validation Loss: 19.025634 Training Accuracy: 0.758929 Validation Accuracy: 0.745201 Validation loss decreased (inf --> 19.025634). Saving model ...
Epoch: 2 Training Loss: 8.630891 Validation Loss: 11.952998 Training Accuracy: 0.875149 Validation Accuracy: 0.812454 Validation loss decreased (19.025634 --> 11.952998). Saving model ...
Epoch: 3 Training Loss: 6.160637 Validation Loss: 5.480922 Training Accuracy: 0.921905 Validation Accuracy: 0.919560 Validation loss decreased (11.952998 --> 5.480922). Saving model ...
Epoch: 4 Training Loss: 5.133198 Validation Loss: 4.369609 Training Accuracy: 0.936459 Validation Accuracy: 0.944615 Validation loss decreased (5.480922 --> 4.369609). Saving model ...
Epoch: 5 Training Loss: 4.771325 Validation Loss: 4.146960 Training Accuracy: 0.941727 Validation Accuracy: 0.947985 Validation loss decreased (4.369609 --> 4.146960). Saving model ...
Epoch: 6 Training Loss: 4.468520 Validation Loss: 4.065775 Training Accuracy: 0.945477 Validation Accuracy: 0.954286 Validation loss decreased (4.146960 --> 4.065775). Saving model ...
Epoch: 7 Training Loss: 4.318840 Validation Loss: 4.137589 Training Accuracy: 0.947917 Validation Accuracy: 0.949377
Epoch: 8 Training Loss: 4.236315 Validation Loss: 4.223764 Training Accuracy: 0.949762 Validation Accuracy: 0.943809
Epoch: 9 Training Loss: 4.423145 Validation Loss: 4.090816 Training Accuracy: 0.946459 Validation Accuracy: 0.943810
Epoch: 10 Training Loss: 4.370817 Validation Loss: 4.102598 Training Accuracy: 0.947857 Validation Accuracy: 0.950330
Epoch: 11 Training Loss: 4.425606 Validation Loss: 4.342477 Training Accuracy: 0.942441 Validation Accuracy: 0.943516
Epoch: 12 Training Loss: 4.373963 Validation Loss: 4.197989 Training Accuracy: 0.947143 Validation Accuracy: 0.946374
Epoch: 13 Training Loss: 4.505095 Validation Loss: 3.972159 Training Accuracy: 0.947917 Validation Accuracy: 0.946667 Validation loss decreased (4.065775 --> 3.972159). Saving model ...
Epoch: 14 Training Loss: 4.450557 Validation Loss: 4.139354 Training Accuracy: 0.947173 Validation Accuracy: 0.945421
Epoch: 15 Training Loss: 4.492465 Validation Loss: 4.199612 Training Accuracy: 0.947143 Validation Accuracy: 0.941465
Epoch: 16 Training Loss: 4.328222 Validation Loss: 4.019412 Training Accuracy: 0.949078 Validation Accuracy: 0.947473
Epoch: 17 Training Loss: 4.180741 Validation Loss: 4.101105 Training Accuracy: 0.951221 Validation Accuracy: 0.943223
Epoch: 18 Training Loss: 4.424772 Validation Loss: 4.231482 Training Accuracy: 0.945477 Validation Accuracy: 0.945714
Epoch: 19 Training Loss: 4.411623 Validation Loss: 4.089917 Training Accuracy: 0.947679 Validation Accuracy: 0.949524
Epoch: 20 Training Loss: 4.331415 Validation Loss: 4.165735 Training Accuracy: 0.945238 Validation Accuracy: 0.945421
Epoch: 21 Training Loss: 4.366600 Validation Loss: 3.887424 Training Accuracy: 0.949078 Validation Accuracy: 0.947473 Validation loss decreased (3.972159 --> 3.887424). Saving model ...
Epoch: 22 Training Loss: 4.308535 Validation Loss: 4.079916 Training Accuracy: 0.950536 Validation Accuracy: 0.950330
Epoch: 23 Training Loss: 4.471057 Validation Loss: 4.255373 Training Accuracy: 0.945982 Validation Accuracy: 0.944469
Epoch: 24 Training Loss: 4.319843 Validation Loss: 4.162312 Training Accuracy: 0.947857 Validation Accuracy: 0.945568
Epoch: 25 Training Loss: 4.445450 Validation Loss: 3.934991 Training Accuracy: 0.948602 Validation Accuracy: 0.951428
Epoch: 26 Training Loss: 4.202756 Validation Loss: 4.202244 Training Accuracy: 0.951488 Validation Accuracy: 0.944176
Epoch: 27 Training Loss: 4.362491 Validation Loss: 3.983603 Training Accuracy: 0.947411 Validation Accuracy: 0.952235
Epoch: 28 Training Loss: 4.465144 Validation Loss: 4.075760 Training Accuracy: 0.947173 Validation Accuracy: 0.944615
Epoch: 29 Training Loss: 4.260040 Validation Loss: 4.003717 Training Accuracy: 0.951667 Validation Accuracy: 0.951135
Epoch: 30 Training Loss: 4.308027 Validation Loss: 3.967594 Training Accuracy: 0.949524 Validation Accuracy: 0.949524
Epoch: 31 Training Loss: 4.412467 Validation Loss: 4.120583 Training Accuracy: 0.950238 Validation Accuracy: 0.945568
Epoch: 32 Training Loss: 4.306862 Validation Loss: 4.055112 Training Accuracy: 0.944286 Validation Accuracy: 0.943663
Epoch: 33 Training Loss: 4.203272 Validation Loss: 3.992813 Training Accuracy: 0.954286 Validation Accuracy: 0.952234
Epoch: 34 Training Loss: 4.476393 Validation Loss: 4.119056 Training Accuracy: 0.947381 Validation Accuracy: 0.949084
Epoch: 35 Training Loss: 4.403752 Validation Loss: 4.100907 Training Accuracy: 0.945000 Validation Accuracy: 0.948278
Epoch: 36 Training Loss: 4.105152 Validation Loss: 4.132325 Training Accuracy: 0.954554 Validation Accuracy: 0.945568
Epoch: 37 Training Loss: 4.297418 Validation Loss: 4.272868 Training Accuracy: 0.947411 Validation Accuracy: 0.940806
Epoch: 38 Training Loss: 4.441558 Validation Loss: 3.982885 Training Accuracy: 0.949792 Validation Accuracy: 0.949231
Epoch: 39 Training Loss: 4.327732 Validation Loss: 4.004139 Training Accuracy: 0.952857 Validation Accuracy: 0.946667
Epoch: 40 Training Loss: 4.248325 Validation Loss: 3.877483 Training Accuracy: 0.950268 Validation Accuracy: 0.947326 Validation loss decreased (3.887424 --> 3.877483). Saving model ...
Epoch: 41 Training Loss: 4.471760 Validation Loss: 4.095389 Training Accuracy: 0.946727 Validation Accuracy: 0.948571
Epoch: 42 Training Loss: 4.514648 Validation Loss: 3.955327 Training Accuracy: 0.946191 Validation Accuracy: 0.944615
Epoch: 43 Training Loss: 4.235168 Validation Loss: 3.982700 Training Accuracy: 0.948602 Validation Accuracy: 0.951282
Epoch: 44 Training Loss: 4.467456 Validation Loss: 4.040695 Training Accuracy: 0.946459 Validation Accuracy: 0.949377
Epoch: 45 Training Loss: 4.323918 Validation Loss: 3.881235 Training Accuracy: 0.950477 Validation Accuracy: 0.950330
Epoch: 46 Training Loss: 4.229419 Validation Loss: 4.030615 Training Accuracy: 0.949078 Validation Accuracy: 0.952381
Epoch: 47 Training Loss: 4.589242 Validation Loss: 4.148905 Training Accuracy: 0.945745 Validation Accuracy: 0.940952
Epoch: 48 Training Loss: 4.355665 Validation Loss: 4.057556 Training Accuracy: 0.946459 Validation Accuracy: 0.946227
Epoch: 49 Training Loss: 4.616849 Validation Loss: 4.143768 Training Accuracy: 0.945268 Validation Accuracy: 0.941758
Epoch: 50 Training Loss: 4.157274 Validation Loss: 4.051403 Training Accuracy: 0.948155 Validation Accuracy: 0.948425
Epoch: 51 Training Loss: 4.330820 Validation Loss: 3.976135 Training Accuracy: 0.946727 Validation Accuracy: 0.943663
Epoch: 52 Training Loss: 4.509131 Validation Loss: 4.210720 Training Accuracy: 0.947441 Validation Accuracy: 0.940806
Epoch: 53 Training Loss: 4.363433 Validation Loss: 4.085523 Training Accuracy: 0.946310 Validation Accuracy: 0.942711
Epoch: 54 Training Loss: 4.329374 Validation Loss: 4.128520 Training Accuracy: 0.946756 Validation Accuracy: 0.946520
Epoch: 55 Training Loss: 4.400236 Validation Loss: 3.954015 Training Accuracy: 0.949048 Validation Accuracy: 0.952088
Epoch: 56 Training Loss: 4.504717 Validation Loss: 4.235549 Training Accuracy: 0.948572 Validation Accuracy: 0.946080
Epoch: 57 Training Loss: 4.463024 Validation Loss: 3.948046 Training Accuracy: 0.943125 Validation Accuracy: 0.947619
Epoch: 58 Training Loss: 4.398079 Validation Loss: 3.897638 Training Accuracy: 0.946667 Validation Accuracy: 0.948278
Epoch: 59 Training Loss: 4.376378 Validation Loss: 4.081883 Training Accuracy: 0.949078 Validation Accuracy: 0.945421
Epoch: 60 Training Loss: 4.291211 Validation Loss: 4.113515 Training Accuracy: 0.949048 Validation Accuracy: 0.949524
Epoch: 61 Training Loss: 4.572069 Validation Loss: 4.129477 Training Accuracy: 0.946429 Validation Accuracy: 0.949377
Epoch: 62 Training Loss: 4.531359 Validation Loss: 4.047833 Training Accuracy: 0.946191 Validation Accuracy: 0.949377
Epoch: 63 Training Loss: 4.382372 Validation Loss: 4.225421 Training Accuracy: 0.947709 Validation Accuracy: 0.942711
Epoch: 64 Training Loss: 4.147504 Validation Loss: 4.304519 Training Accuracy: 0.947203 Validation Accuracy: 0.941905
Epoch: 65 Training Loss: 4.354286 Validation Loss: 4.034191 Training Accuracy: 0.949792 Validation Accuracy: 0.952234
Epoch: 66 Training Loss: 4.615275 Validation Loss: 4.292393 Training Accuracy: 0.944762 Validation Accuracy: 0.944469
Epoch: 67 Training Loss: 4.416802 Validation Loss: 4.154194 Training Accuracy: 0.947203 Validation Accuracy: 0.947473
Epoch: 68 Training Loss: 4.463829 Validation Loss: 4.224046 Training Accuracy: 0.949048 Validation Accuracy: 0.943809
Epoch: 69 Training Loss: 4.397315 Validation Loss: 3.923909 Training Accuracy: 0.947887 Validation Accuracy: 0.949084
Epoch: 70 Training Loss: 4.449658 Validation Loss: 3.949501 Training Accuracy: 0.947381 Validation Accuracy: 0.947473
Epoch: 71 Training Loss: 4.487672 Validation Loss: 3.959450 Training Accuracy: 0.947858 Validation Accuracy: 0.950330
Epoch: 72 Training Loss: 4.458922 Validation Loss: 4.124105 Training Accuracy: 0.944792 Validation Accuracy: 0.949230
Epoch: 73 Training Loss: 4.467346 Validation Loss: 4.053494 Training Accuracy: 0.950268 Validation Accuracy: 0.951282
Epoch: 74 Training Loss: 4.438444 Validation Loss: 4.088782 Training Accuracy: 0.945774 Validation Accuracy: 0.949524
Epoch: 75 Training Loss: 4.458971 Validation Loss: 4.138270 Training Accuracy: 0.947381 Validation Accuracy: 0.950183
Epoch: 76 Training Loss: 4.280409 Validation Loss: 4.181010 Training Accuracy: 0.949316 Validation Accuracy: 0.945714
Epoch: 77 Training Loss: 4.537152 Validation Loss: 4.025983 Training Accuracy: 0.948572 Validation Accuracy: 0.948938
Epoch: 78 Training Loss: 4.424292 Validation Loss: 4.012586 Training Accuracy: 0.946221 Validation Accuracy: 0.956190
Epoch: 79 Training Loss: 4.596270 Validation Loss: 4.072958 Training Accuracy: 0.950268 Validation Accuracy: 0.943809
Epoch: 80 Training Loss: 4.437507 Validation Loss: 3.956301 Training Accuracy: 0.948602 Validation Accuracy: 0.949377
Epoch: 81 Training Loss: 4.719672 Validation Loss: 4.221606 Training Accuracy: 0.941905 Validation Accuracy: 0.945275
Epoch: 82 Training Loss: 4.453207 Validation Loss: 4.037356 Training Accuracy: 0.946697 Validation Accuracy: 0.941905
Epoch: 83 Training Loss: 4.319076 Validation Loss: 4.163154 Training Accuracy: 0.950477 Validation Accuracy: 0.945714
Epoch: 84 Training Loss: 4.506349 Validation Loss: 4.076897 Training Accuracy: 0.946459 Validation Accuracy: 0.941758
Epoch: 85 Training Loss: 4.561115 Validation Loss: 4.265839 Training Accuracy: 0.944792 Validation Accuracy: 0.944762
Epoch: 86 Training Loss: 4.459120 Validation Loss: 4.360689 Training Accuracy: 0.949286 Validation Accuracy: 0.945568
Epoch: 87 Training Loss: 4.599127 Validation Loss: 4.161524 Training Accuracy: 0.946727 Validation Accuracy: 0.946373
Epoch: 88 Training Loss: 4.443397 Validation Loss: 4.160495 Training Accuracy: 0.945834 Validation Accuracy: 0.947473
Epoch: 89 Training Loss: 4.776461 Validation Loss: 3.965901 Training Accuracy: 0.945477 Validation Accuracy: 0.946374
Epoch: 90 Training Loss: 4.400144 Validation Loss: 4.052342 Training Accuracy: 0.945477 Validation Accuracy: 0.946667
Epoch: 91 Training Loss: 4.218872 Validation Loss: 4.026609 Training Accuracy: 0.950268 Validation Accuracy: 0.948571
Epoch: 92 Training Loss: 4.606010 Validation Loss: 3.898826 Training Accuracy: 0.945477 Validation Accuracy: 0.948571
Epoch: 93 Training Loss: 4.566362 Validation Loss: 4.057926 Training Accuracy: 0.945030 Validation Accuracy: 0.946374
Epoch: 94 Training Loss: 4.396929 Validation Loss: 4.165209 Training Accuracy: 0.948572 Validation Accuracy: 0.947472
Epoch: 95 Training Loss: 4.440193 Validation Loss: 4.040040 Training Accuracy: 0.948572 Validation Accuracy: 0.953333
Epoch: 96 Training Loss: 4.378953 Validation Loss: 3.938155 Training Accuracy: 0.945774 Validation Accuracy: 0.946520
Epoch: 97 Training Loss: 4.351010 Validation Loss: 4.056520 Training Accuracy: 0.948572 Validation Accuracy: 0.943516
Epoch: 98 Training Loss: 4.803754 Validation Loss: 4.003698 Training Accuracy: 0.943810 Validation Accuracy: 0.950476
Epoch: 99 Training Loss: 4.442692 Validation Loss: 4.072057 Training Accuracy: 0.947649 Validation Accuracy: 0.947179
Epoch: 100 Training Loss: 4.184012 Validation Loss: 4.058261 Training Accuracy: 0.949108 Validation Accuracy: 0.950330
plt.figure(figsize=(10,8))
plt.plot(train_losses,'--',linewidth = 2.5, label='Training loss')
plt.plot(valid_losses,'--', linewidth = 2.5, label='Validation loss')
plt.xlabel('Epochs',fontsize = 20)
plt.ylim(2.5,20)
plt.legend(fontsize = 14)
plt.xticks(fontsize = 12)
plt.yticks(fontsize = 12);
plt.figure(figsize=(10,8))
plt.plot(train_acc,'--', linewidth = 2.5, label='Training accuracy')
plt.plot(valid_acc,'--', linewidth = 2.5, label='Validation accuracy')
plt.xlabel('Epochs', fontsize = 20)
plt.legend(fontsize = 14)
plt.xticks(fontsize = 12)
plt.yticks(fontsize = 12);
pwd
'/home/sysadm/Documents/3ml_kkm/env'
model.load_state_dict(torch.load('./../models/model_pneumonia.pt'))
<All keys matched successfully>
# tracking test loss
test_loss = 0.0
class_correct = list(0. for i in range(2))
class_total = list(0. for i in range(2))
model.eval()
# iterating over test data
for data, target in tqdm(test_loader, desc = 'Model Test: '):
# moving tensors to GPU if CUDA is available
data = torch.Tensor(data)
target = tuple([0 if i=='NORMAL' else 1 for i in list(target)])
target = torch.Tensor(np.array(target))
if train_on_gpu:
data, target = data.cuda(), target.cuda()
# forward pass: computing predicted outputs by passing inputs to the model
output = model(data)
target = target.long()
# calculating the batch loss
loss = criterion(output, target)
# updating test loss
test_loss += loss.item()*data.size(0)
# converting output probabilities to predicted class
_, pred = torch.max(output, 1)
# comparing predictions to true label
correct_tensor = pred.eq(target.data.view_as(pred))
correct = np.squeeze(correct_tensor.numpy()) if not train_on_gpu else np.squeeze(correct_tensor.cpu().numpy())
# calculating test accuracy for each object class
for i in range(len(target.data)):
label = target.data[i]
class_correct[label] += correct[i].item()
class_total[label] += 1
# average test loss
test_loss = test_loss/len(test_loader.dataset)
print('Test Loss: {:.6f}\n'.format(test_loss))
for i in range(2):
if class_total[i] > 0:
print('Test Accuracy of %5s: %2d%% (%2d/%2d)' % (
classes[i], 100 * class_correct[i] / class_total[i],
np.sum(class_correct[i]), np.sum(class_total[i])))
else:
print('Test Accuracy of %5s: N/A (no training examples)' % (classes[i]))
print('\nTest Accuracy (Overall): %2d%% (%2d/%2d)' % (
100. * np.sum(class_correct) / np.sum(class_total),
np.sum(class_correct), np.sum(class_total)))
Test Loss: 0.445040 Test Accuracy of PNEUMONIA: 61% (143/234) Test Accuracy of NORMAL: 96% (377/390) Test Accuracy (Overall): 83% (520/624)
confusion_matrix = torch.zeros(2,2)
with torch.no_grad():
for i, (data, target) in enumerate(test_loader):
data = torch.Tensor(data)
target = tuple([0 if i=='NORMAL' else 1 for i in list(target)])
target = torch.Tensor(np.array(target))
if train_on_gpu:
data, target = data.cuda(), target.cuda()
output = model(data)
target = target.long()
_, preds = torch.max(output, 1)
for t, p in zip(target.view(-1), preds.view(-1)):
confusion_matrix[t.long(), p.long()] += 1
print(confusion_matrix)
tensor([[143., 91.],
[ 13., 377.]])
cm = confusion_matrix.numpy()
tn, fp, fn, tp = cm.ravel()
precision = tp/(tp+fp)
recall = tp/(tp+fn)
f1_score = (2*precision*recall)/(precision+recall)
print("Precision: {:.4f}\nRecall: {:.4f}\nF1 Score: {:.4f}".format(precision,recall,f1_score))
Precision: 0.8056 Recall: 0.9667 F1 Score: 0.8788
plt.figure(figsize=(8,7))
xticks = np.array(['NORMAL','PNEUMONIA'])
sns.heatmap(cm, annot = True,
xticklabels = xticks, yticklabels = xticks)
plt.xlabel('PREDICTED',fontsize = 15)
plt.ylabel('TRUE',fontsize = 15);
# obtaining one batch of test images
data, target = next(iter(test_loader))
data.numpy()
# moving model inputs to cuda, if GPU available
if train_on_gpu:
data = data.cuda()
# getting sample outputs
output = model(data)
# converting output probabilities to predicted class
_, preds_tensor = torch.max(output, 1)
preds = np.squeeze(preds_tensor.numpy()) if not train_on_gpu else np.squeeze(preds_tensor.cpu().numpy())
target = tuple([0 if i=='NORMAL' else 1 for i in list(target)])
# plotting the images in the batch, along with predicted and true labels
fig = plt.figure(figsize=(20,10))
for idx in np.arange(8):
ax = fig.add_subplot(2,4, idx+1, xticks=[], yticks=[])
ax.imshow(data[idx].permute(1, 2, 0) if not train_on_gpu else data[idx].cpu().permute(1, 2, 0))
ax.set_title("{} ({})".format(classes[preds[idx]], classes[target[idx]]),
color=("green" if preds[idx]==target[idx] else "red"))
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
train_transforms_res = transforms.Compose([transforms.RandomRotation(30),
transforms.Resize((224,224)),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize((0.485, 0.456, 0.406),(0.229, 0.224, 0.225))
])
test_transforms_res = transforms.Compose([transforms.Resize((224,224)),
transforms.ToTensor(),
transforms.Normalize((0.485, 0.456, 0.406),(0.229, 0.224, 0.225))
])
train_data_res = datasets.ImageFolder(train_dir, transform = train_transforms_res)
test_data_res = datasets.ImageFolder(test_dir, transform = test_transforms_res)
# training indices to be used for validation
num_train_res = len(train_data_res); print(num_train_res)
indices = list(range(num_train_res))
np.random.shuffle(indices)
split = int(np.floor(0.2 * num_train_res)); print(split)
train_idx, valid_idx = indices[split:], indices[:split]
5232 1046
# defining samplers for obtaining training and validation batches
train_sampler = SubsetRandomSampler(train_idx)
valid_sampler = SubsetRandomSampler(valid_idx)
# loading the data
train_loader = torch.utils.data.DataLoader(train_data_res, batch_size = 30, sampler = train_sampler)
valid_loader = torch.utils.data.DataLoader(train_data_res, batch_size = 30, sampler = valid_sampler)
test_loader = torch.utils.data.DataLoader(test_data_res, batch_size = 30, shuffle = True)
next(iter(test_loader))
[tensor([[[[-0.9534, -0.9020, -0.8678, ..., -1.3644, -1.3815, -1.4500],
[-0.9020, -0.9020, -0.8849, ..., -1.3473, -1.3815, -1.4500],
[-0.9192, -0.9192, -0.9020, ..., -1.3302, -1.3815, -1.4500],
...,
[-1.7412, -1.7412, -1.7583, ..., -1.7583, -1.7754, -1.7583],
[-1.7412, -1.7412, -1.7583, ..., -1.7583, -1.7754, -1.7925],
[-1.7412, -1.7412, -1.7583, ..., -1.7412, -1.7583, -1.7925]],
[[-0.8452, -0.7927, -0.7577, ..., -1.2654, -1.2829, -1.3529],
[-0.7927, -0.7927, -0.7752, ..., -1.2479, -1.2829, -1.3529],
[-0.8102, -0.8102, -0.7927, ..., -1.2304, -1.2829, -1.3529],
...,
[-1.6506, -1.6506, -1.6681, ..., -1.6681, -1.6856, -1.6681],
[-1.6506, -1.6506, -1.6681, ..., -1.6681, -1.6856, -1.7031],
[-1.6506, -1.6506, -1.6681, ..., -1.6506, -1.6681, -1.7031]],
[[-0.6193, -0.5670, -0.5321, ..., -1.0376, -1.0550, -1.1247],
[-0.5670, -0.5670, -0.5495, ..., -1.0201, -1.0550, -1.1247],
[-0.5844, -0.5844, -0.5670, ..., -1.0027, -1.0550, -1.1247],
...,
[-1.4210, -1.4210, -1.4384, ..., -1.4384, -1.4559, -1.4384],
[-1.4210, -1.4210, -1.4384, ..., -1.4384, -1.4559, -1.4733],
[-1.4210, -1.4210, -1.4384, ..., -1.4210, -1.4384, -1.4733]]],
[[[-2.1179, -2.1179, -2.0837, ..., -2.1179, -1.8268, -1.1932],
[-2.1179, -2.1179, -2.1008, ..., -2.1179, -1.8268, -1.1760],
[-2.1179, -2.1179, -2.1008, ..., -2.1008, -1.8097, -1.1760],
...,
[-2.1179, -2.1179, -2.1179, ..., -2.1179, -2.1179, -2.1179],
[-2.1179, -2.1179, -2.1179, ..., -2.1179, -2.1179, -2.1179],
[-2.1179, -2.1179, -2.1179, ..., -2.1179, -2.1179, -2.1179]],
[[-2.0357, -2.0357, -2.0007, ..., -2.0357, -1.7381, -1.0903],
[-2.0357, -2.0357, -2.0182, ..., -2.0357, -1.7381, -1.0728],
[-2.0357, -2.0357, -2.0182, ..., -2.0182, -1.7206, -1.0728],
...,
[-2.0357, -2.0357, -2.0357, ..., -2.0357, -2.0357, -2.0357],
[-2.0357, -2.0357, -2.0357, ..., -2.0357, -2.0357, -2.0357],
[-2.0357, -2.0357, -2.0357, ..., -2.0357, -2.0357, -2.0357]],
[[-1.8044, -1.8044, -1.7696, ..., -1.8044, -1.5081, -0.8633],
[-1.8044, -1.8044, -1.7870, ..., -1.8044, -1.5081, -0.8458],
[-1.8044, -1.8044, -1.7870, ..., -1.7870, -1.4907, -0.8458],
...,
[-1.8044, -1.8044, -1.8044, ..., -1.8044, -1.8044, -1.8044],
[-1.8044, -1.8044, -1.8044, ..., -1.8044, -1.8044, -1.8044],
[-1.8044, -1.8044, -1.8044, ..., -1.8044, -1.8044, -1.8044]]],
[[[-0.9363, -0.7479, -0.6109, ..., 1.1872, 1.8208, 2.1975],
[-0.9192, -0.7137, -0.5938, ..., 1.1872, 1.8037, 2.1975],
[-0.9192, -0.6965, -0.5767, ..., 1.1529, 1.7694, 2.1804],
...,
[-2.1179, -2.1179, -2.1179, ..., -2.1179, -2.1179, -1.9467],
[-2.1179, -2.1179, -2.1179, ..., -2.1179, -2.1179, -1.9638],
[-2.1179, -2.1179, -2.1179, ..., -2.1179, -2.1179, -1.9467]],
[[-0.8277, -0.6352, -0.4951, ..., 1.3431, 1.9909, 2.3761],
[-0.8102, -0.6001, -0.4776, ..., 1.3431, 1.9734, 2.3761],
[-0.8102, -0.5826, -0.4601, ..., 1.3081, 1.9384, 2.3585],
...,
[-2.0357, -2.0357, -2.0357, ..., -2.0357, -2.0357, -1.8606],
[-2.0357, -2.0357, -2.0357, ..., -2.0357, -2.0357, -1.8782],
[-2.0357, -2.0357, -2.0357, ..., -2.0357, -2.0357, -1.8606]],
[[-0.6018, -0.4101, -0.2707, ..., 1.5594, 2.2043, 2.5877],
[-0.5844, -0.3753, -0.2532, ..., 1.5594, 2.1868, 2.5877],
[-0.5844, -0.3578, -0.2358, ..., 1.5245, 2.1520, 2.5703],
...,
[-1.8044, -1.8044, -1.8044, ..., -1.8044, -1.8044, -1.6302],
[-1.8044, -1.8044, -1.8044, ..., -1.8044, -1.8044, -1.6476],
[-1.8044, -1.8044, -1.8044, ..., -1.8044, -1.8044, -1.6302]]],
...,
[[[-2.1179, -2.1179, -2.1179, ..., -2.1179, -2.1179, -2.1179],
[-2.1179, -2.1179, -2.1179, ..., -2.1179, -2.1179, -2.1179],
[-2.1179, -2.1179, -2.1179, ..., -2.1179, -2.1179, -2.1179],
...,
[-2.0665, -1.8439, -1.6384, ..., -2.1179, -2.1179, -2.1179],
[-2.0323, -1.7754, -1.5870, ..., -2.1179, -2.1179, -2.1179],
[-2.0323, -1.7412, -1.5014, ..., -2.1179, -2.1179, -2.1179]],
[[-2.0357, -2.0357, -2.0357, ..., -2.0357, -2.0357, -2.0357],
[-2.0357, -2.0357, -2.0357, ..., -2.0357, -2.0357, -2.0357],
[-2.0357, -2.0357, -2.0357, ..., -2.0357, -2.0357, -2.0357],
...,
[-1.9832, -1.7556, -1.5455, ..., -2.0357, -2.0357, -2.0357],
[-1.9482, -1.6856, -1.4930, ..., -2.0357, -2.0357, -2.0357],
[-1.9482, -1.6506, -1.4055, ..., -2.0357, -2.0357, -2.0357]],
[[-1.8044, -1.8044, -1.8044, ..., -1.8044, -1.8044, -1.8044],
[-1.8044, -1.8044, -1.8044, ..., -1.8044, -1.8044, -1.8044],
[-1.8044, -1.8044, -1.8044, ..., -1.8044, -1.8044, -1.8044],
...,
[-1.7522, -1.5256, -1.3164, ..., -1.8044, -1.8044, -1.8044],
[-1.7173, -1.4559, -1.2641, ..., -1.8044, -1.8044, -1.8044],
[-1.7173, -1.4210, -1.1770, ..., -1.8044, -1.8044, -1.8044]]],
[[[-1.3815, -1.3987, -1.3473, ..., 0.5536, 0.7077, 0.8447],
[-1.3644, -1.3987, -1.3473, ..., 0.5878, 0.7933, 0.7933],
[-1.3302, -1.3644, -1.3473, ..., 0.5878, 0.6734, 0.6563],
...,
[-1.8953, -1.8953, -1.9638, ..., -1.6555, -1.5528, -1.3644],
[-1.8782, -1.9124, -1.9809, ..., -1.6555, -1.5357, -1.3644],
[-1.8782, -1.9124, -1.9809, ..., -1.6384, -1.5357, -1.3302]],
[[-1.2829, -1.3004, -1.2479, ..., 0.6954, 0.8529, 0.9930],
[-1.2654, -1.3004, -1.2479, ..., 0.7304, 0.9405, 0.9405],
[-1.2304, -1.2654, -1.2479, ..., 0.7304, 0.8179, 0.8004],
...,
[-1.8081, -1.8081, -1.8782, ..., -1.5630, -1.4580, -1.2654],
[-1.7906, -1.8256, -1.8957, ..., -1.5630, -1.4405, -1.2654],
[-1.7906, -1.8256, -1.8957, ..., -1.5455, -1.4405, -1.2304]],
[[-1.0550, -1.0724, -1.0201, ..., 0.9145, 1.0714, 1.2108],
[-1.0376, -1.0724, -1.0201, ..., 0.9494, 1.1585, 1.1585],
[-1.0027, -1.0376, -1.0201, ..., 0.9494, 1.0365, 1.0191],
...,
[-1.5779, -1.5779, -1.6476, ..., -1.3339, -1.2293, -1.0376],
[-1.5604, -1.5953, -1.6650, ..., -1.3339, -1.2119, -1.0376],
[-1.5604, -1.5953, -1.6650, ..., -1.3164, -1.2119, -1.0027]]],
[[[-0.4054, -0.6794, -1.1589, ..., -1.5357, -1.5357, -1.5357],
[-0.3369, -0.7479, -1.1932, ..., -1.5528, -1.5528, -1.5185],
[-0.4911, -0.9192, -1.1760, ..., -1.5185, -1.5357, -1.5357],
...,
[-1.6898, -1.4500, -1.2959, ..., -1.9809, -1.9295, -1.8439],
[-1.6898, -1.4672, -1.3130, ..., -1.9809, -1.9295, -1.8439],
[-1.6898, -1.4500, -1.2788, ..., -1.9809, -1.9295, -1.8439]],
[[-0.2850, -0.5651, -1.0553, ..., -1.4405, -1.4405, -1.4405],
[-0.2150, -0.6352, -1.0903, ..., -1.4580, -1.4580, -1.4230],
[-0.3725, -0.8102, -1.0728, ..., -1.4230, -1.4405, -1.4405],
...,
[-1.5980, -1.3529, -1.1954, ..., -1.8957, -1.8431, -1.7556],
[-1.5980, -1.3704, -1.2129, ..., -1.8957, -1.8431, -1.7556],
[-1.5980, -1.3529, -1.1779, ..., -1.8957, -1.8431, -1.7556]],
[[-0.0615, -0.3404, -0.8284, ..., -1.2119, -1.2119, -1.2119],
[ 0.0082, -0.4101, -0.8633, ..., -1.2293, -1.2293, -1.1944],
[-0.1487, -0.5844, -0.8458, ..., -1.1944, -1.2119, -1.2119],
...,
[-1.3687, -1.1247, -0.9678, ..., -1.6650, -1.6127, -1.5256],
[-1.3687, -1.1421, -0.9853, ..., -1.6650, -1.6127, -1.5256],
[-1.3687, -1.1247, -0.9504, ..., -1.6650, -1.6127, -1.5256]]]]),
tensor([1, 0, 0, 0, 1, 0, 1, 0, 1, 1, 1, 0, 0, 1, 0, 1, 1, 1, 1, 1, 0, 1, 1, 0,
1, 1, 0, 0, 1, 1])]
show_images(test_loader)
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
res_model = models.resnet50(pretrained=True)
print(res_model)
ResNet(
(conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(maxpool): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)
(layer1): Sequential(
(0): Bottleneck(
(conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Bottleneck(
(conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): Bottleneck(
(conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(layer2): Sequential(
(0): Bottleneck(
(conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Bottleneck(
(conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): Bottleneck(
(conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(3): Bottleneck(
(conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(layer3): Sequential(
(0): Bottleneck(
(conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(512, 1024, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(3): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(4): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(5): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(layer4): Sequential(
(0): Bottleneck(
(conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Bottleneck(
(conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): Bottleneck(
(conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(avgpool): AdaptiveAvgPool2d(output_size=(1, 1))
(fc): Linear(in_features=2048, out_features=1000, bias=True)
)
# Freezing training for all "features" layers
for param in res_model.parameters():
param.requires_grad = False
# Adding final layers
res_model.fc = nn.Sequential(nn.Linear(2048,512),
nn.BatchNorm1d(512),
nn.ReLU(),
nn.Dropout(0.4),
nn.Linear(512,2),
nn.LogSoftmax(dim=1))
# specifying the loss function (categorical cross-entropy)
criterion = nn.CrossEntropyLoss()
# specifying optimizer
optimizer = optim.SGD(res_model.parameters(), lr = 0.008,momentum = 0.9, nesterov = True)
# specifying scheduler
scheduler = optim.lr_scheduler.ExponentialLR(optimizer,gamma = 0.2)
if train_on_gpu:
res_model.cuda()
print(res_model)
ResNet(
(conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(maxpool): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)
(layer1): Sequential(
(0): Bottleneck(
(conv1): Conv2d(64, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Bottleneck(
(conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): Bottleneck(
(conv1): Conv2d(256, 64, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(64, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(layer2): Sequential(
(0): Bottleneck(
(conv1): Conv2d(256, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Bottleneck(
(conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): Bottleneck(
(conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(3): Bottleneck(
(conv1): Conv2d(512, 128, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(128, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(layer3): Sequential(
(0): Bottleneck(
(conv1): Conv2d(512, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(512, 1024, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(3): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(4): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(5): Bottleneck(
(conv1): Conv2d(1024, 256, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(256, 1024, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(layer4): Sequential(
(0): Bottleneck(
(conv1): Conv2d(1024, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
(downsample): Sequential(
(0): Conv2d(1024, 2048, kernel_size=(1, 1), stride=(2, 2), bias=False)
(1): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
)
)
(1): Bottleneck(
(conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
(2): Bottleneck(
(conv1): Conv2d(2048, 512, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)
(bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(conv3): Conv2d(512, 2048, kernel_size=(1, 1), stride=(1, 1), bias=False)
(bn3): BatchNorm2d(2048, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(relu): ReLU(inplace=True)
)
)
(avgpool): AdaptiveAvgPool2d(output_size=(1, 1))
(fc): Sequential(
(0): Linear(in_features=2048, out_features=512, bias=True)
(1): BatchNorm1d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
(3): Dropout(p=0.4, inplace=False)
(4): Linear(in_features=512, out_features=2, bias=True)
(5): LogSoftmax(dim=1)
)
)
# number of epochs to train the model
n_epochs = 100
valid_loss_min = np.Inf # track change in validation loss
train_losses, valid_losses = [], []
train_acc, valid_acc = [], []
for epoch in tqdm(range(1, n_epochs+1), desc = 'Model Train/Epoch: '):
# keeping track of loss and accuracy
train_loss = 0.0
valid_loss = 0.0
train_accuracy = 0.0
valid_accuracy = 0.0
######################
# training the model #
######################
res_model.train()
for data, target in tqdm(train_loader, desc = 'Model Train: '):
# moving tensors to GPU if CUDA is available
if train_on_gpu:
data, target = data.cuda(), target.cuda()
# clearing the gradients of all optimized variables
optimizer.zero_grad()
# forward pass: computing predicted outputs by passing inputs to the model
output = res_model(data)
target = target.long()
# calculating the batch loss
loss = criterion(output, target)
# backward pass: computing gradient of the loss with respect to model parameters
loss.backward()
# performing a single optimization step (parameter update)
optimizer.step()
# updating training loss
train_loss += loss.item()*data.size(0)
# updating training accuracy
train_accuracy += accuracy(output,target)
scheduler.step() # decay LR
########################
# validating the model #
########################
res_model.eval()
for data, target in tqdm(valid_loader, desc = 'Model Validation: '):
# moving tensors to GPU if CUDA is available
if train_on_gpu:
data, target = data.cuda(), target.cuda()
# forward pass: computing predicted outputs by passing inputs to the model
output = res_model(data)
target = target.long()
# calculating the batch loss
loss = criterion(output, target)
# updating average validation loss
valid_loss += loss.item()*data.size(0)
# updating validation accuracy
valid_accuracy += accuracy(output,target)
# calculating average losses
train_loss = train_loss/len(train_loader)
valid_loss = valid_loss/len(valid_loader)
# calculating average accuracy
train_accuracy = train_accuracy/len(train_loader)
valid_accuracy = valid_accuracy/len(valid_loader)
train_losses.append(train_loss)
valid_losses.append(valid_loss)
train_acc.append(train_accuracy)
valid_acc.append(valid_accuracy)
# printing training/validation statistics
print('Epoch: {} \tTraining Loss: {:.6f} \tValidation Loss: {:.6f} \n\t\tTraining Accuracy: {:.6f} \tValidation Accuracy: {:.6f}'.format(
epoch, train_loss, valid_loss, train_accuracy, valid_accuracy))
# saving model if validation loss has decreased
if valid_loss <= valid_loss_min:
print('Validation loss decreased ({:.6f} --> {:.6f}). Saving model ...'.format(
valid_loss_min,
valid_loss))
torch.save(res_model.state_dict(), './../models/res_model_pneumonia.pt')
valid_loss_min = valid_loss
Epoch: 1 Training Loss: 5.319068 Validation Loss: 5.963136 Training Accuracy: 0.927411 Validation Accuracy: 0.925714 Validation loss decreased (inf --> 5.963136). Saving model ...
Epoch: 2 Training Loss: 4.056168 Validation Loss: 2.948793 Training Accuracy: 0.947381 Validation Accuracy: 0.960659 Validation loss decreased (5.963136 --> 2.948793). Saving model ...
Epoch: 3 Training Loss: 3.806745 Validation Loss: 3.142063 Training Accuracy: 0.950000 Validation Accuracy: 0.965568
Epoch: 4 Training Loss: 3.629755 Validation Loss: 2.770528 Training Accuracy: 0.952679 Validation Accuracy: 0.969231 Validation loss decreased (2.948793 --> 2.770528). Saving model ...
Epoch: 5 Training Loss: 3.652192 Validation Loss: 2.991458 Training Accuracy: 0.952173 Validation Accuracy: 0.962857
Epoch: 6 Training Loss: 3.432711 Validation Loss: 3.251117 Training Accuracy: 0.955476 Validation Accuracy: 0.958901
Epoch: 7 Training Loss: 3.550301 Validation Loss: 3.362435 Training Accuracy: 0.953185 Validation Accuracy: 0.952088
Epoch: 8 Training Loss: 3.720683 Validation Loss: 2.979840 Training Accuracy: 0.954286 Validation Accuracy: 0.959048
Epoch: 9 Training Loss: 3.465660 Validation Loss: 2.911785 Training Accuracy: 0.954137 Validation Accuracy: 0.959048
Epoch: 10 Training Loss: 3.606291 Validation Loss: 2.476047 Training Accuracy: 0.955030 Validation Accuracy: 0.971282 Validation loss decreased (2.770528 --> 2.476047). Saving model ...
Epoch: 11 Training Loss: 3.558153 Validation Loss: 2.724567 Training Accuracy: 0.954316 Validation Accuracy: 0.965568
Epoch: 12 Training Loss: 3.636773 Validation Loss: 3.364319 Training Accuracy: 0.955000 Validation Accuracy: 0.957949
Epoch: 13 Training Loss: 3.504733 Validation Loss: 3.120922 Training Accuracy: 0.952143 Validation Accuracy: 0.959707
Epoch: 14 Training Loss: 3.436907 Validation Loss: 2.781393 Training Accuracy: 0.952441 Validation Accuracy: 0.965568
Epoch: 15 Training Loss: 3.467195 Validation Loss: 3.190658 Training Accuracy: 0.955566 Validation Accuracy: 0.958095
Epoch: 16 Training Loss: 3.561147 Validation Loss: 2.710242 Training Accuracy: 0.953096 Validation Accuracy: 0.967180
Epoch: 17 Training Loss: 3.394059 Validation Loss: 3.083567 Training Accuracy: 0.954792 Validation Accuracy: 0.967619
Epoch: 18 Training Loss: 3.545196 Validation Loss: 3.295047 Training Accuracy: 0.954554 Validation Accuracy: 0.959048
Epoch: 19 Training Loss: 3.483320 Validation Loss: 3.160184 Training Accuracy: 0.954048 Validation Accuracy: 0.959560
Epoch: 20 Training Loss: 3.753543 Validation Loss: 2.983220 Training Accuracy: 0.952858 Validation Accuracy: 0.968571
Epoch: 21 Training Loss: 3.796485 Validation Loss: 2.773243 Training Accuracy: 0.953572 Validation Accuracy: 0.963663
Epoch: 22 Training Loss: 3.456433 Validation Loss: 2.622279 Training Accuracy: 0.955000 Validation Accuracy: 0.969231
Epoch: 23 Training Loss: 3.569574 Validation Loss: 2.905816 Training Accuracy: 0.953334 Validation Accuracy: 0.962857
Epoch: 24 Training Loss: 3.689424 Validation Loss: 2.975902 Training Accuracy: 0.951697 Validation Accuracy: 0.965568
Epoch: 25 Training Loss: 3.845975 Validation Loss: 2.783165 Training Accuracy: 0.951905 Validation Accuracy: 0.966520
Epoch: 26 Training Loss: 3.527705 Validation Loss: 2.981219 Training Accuracy: 0.951727 Validation Accuracy: 0.961905
Epoch: 27 Training Loss: 3.453213 Validation Loss: 2.799123 Training Accuracy: 0.955238 Validation Accuracy: 0.964762
Epoch: 28 Training Loss: 3.593889 Validation Loss: 2.795605 Training Accuracy: 0.954762 Validation Accuracy: 0.966227
Epoch: 29 Training Loss: 3.738690 Validation Loss: 2.916857 Training Accuracy: 0.950477 Validation Accuracy: 0.962711
Epoch: 30 Training Loss: 3.850638 Validation Loss: 3.134185 Training Accuracy: 0.948869 Validation Accuracy: 0.959560
Epoch: 31 Training Loss: 3.826182 Validation Loss: 2.826899 Training Accuracy: 0.950982 Validation Accuracy: 0.967472
Epoch: 32 Training Loss: 3.646333 Validation Loss: 3.308476 Training Accuracy: 0.953572 Validation Accuracy: 0.961758
Epoch: 33 Training Loss: 3.462745 Validation Loss: 3.023634 Training Accuracy: 0.958095 Validation Accuracy: 0.964469
Epoch: 34 Training Loss: 3.603177 Validation Loss: 2.696172 Training Accuracy: 0.954792 Validation Accuracy: 0.967472
Epoch: 35 Training Loss: 3.579439 Validation Loss: 2.668017 Training Accuracy: 0.952500 Validation Accuracy: 0.964762
Epoch: 36 Training Loss: 3.578393 Validation Loss: 3.347871 Training Accuracy: 0.952173 Validation Accuracy: 0.964762
Epoch: 37 Training Loss: 3.476236 Validation Loss: 3.019542 Training Accuracy: 0.957679 Validation Accuracy: 0.966227
Epoch: 38 Training Loss: 3.769048 Validation Loss: 2.978849 Training Accuracy: 0.946965 Validation Accuracy: 0.962857
Epoch: 39 Training Loss: 3.380495 Validation Loss: 3.127536 Training Accuracy: 0.957619 Validation Accuracy: 0.967619
Epoch: 40 Training Loss: 3.653633 Validation Loss: 3.016873 Training Accuracy: 0.951935 Validation Accuracy: 0.957802
Epoch: 41 Training Loss: 3.418297 Validation Loss: 2.881556 Training Accuracy: 0.953125 Validation Accuracy: 0.960659
Epoch: 42 Training Loss: 3.368543 Validation Loss: 2.932105 Training Accuracy: 0.958334 Validation Accuracy: 0.960000
Epoch: 43 Training Loss: 3.680399 Validation Loss: 2.778480 Training Accuracy: 0.950744 Validation Accuracy: 0.969377
Epoch: 44 Training Loss: 3.347399 Validation Loss: 2.688709 Training Accuracy: 0.956905 Validation Accuracy: 0.970476
Epoch: 45 Training Loss: 3.678187 Validation Loss: 2.939470 Training Accuracy: 0.951250 Validation Accuracy: 0.963663
Epoch: 46 Training Loss: 3.320547 Validation Loss: 2.850448 Training Accuracy: 0.955982 Validation Accuracy: 0.972234
Epoch: 47 Training Loss: 3.419568 Validation Loss: 2.901439 Training Accuracy: 0.956697 Validation Accuracy: 0.960806
Epoch: 48 Training Loss: 3.580173 Validation Loss: 2.727599 Training Accuracy: 0.954524 Validation Accuracy: 0.969524
Epoch: 49 Training Loss: 3.573213 Validation Loss: 2.722480 Training Accuracy: 0.953125 Validation Accuracy: 0.966374
Epoch: 50 Training Loss: 3.631605 Validation Loss: 2.949453 Training Accuracy: 0.952173 Validation Accuracy: 0.958901
Epoch: 51 Training Loss: 3.514179 Validation Loss: 2.734720 Training Accuracy: 0.953155 Validation Accuracy: 0.973333
Epoch: 52 Training Loss: 3.423712 Validation Loss: 3.185203 Training Accuracy: 0.954048 Validation Accuracy: 0.958755
Epoch: 53 Training Loss: 3.526333 Validation Loss: 2.627717 Training Accuracy: 0.955030 Validation Accuracy: 0.971429
Epoch: 54 Training Loss: 3.610644 Validation Loss: 2.882945 Training Accuracy: 0.953363 Validation Accuracy: 0.962418
Epoch: 55 Training Loss: 3.551864 Validation Loss: 2.403123 Training Accuracy: 0.951965 Validation Accuracy: 0.972235 Validation loss decreased (2.476047 --> 2.403123). Saving model ...
Epoch: 56 Training Loss: 3.585871 Validation Loss: 2.514389 Training Accuracy: 0.954286 Validation Accuracy: 0.971428
Epoch: 57 Training Loss: 3.854289 Validation Loss: 2.614783 Training Accuracy: 0.947858 Validation Accuracy: 0.965568
Epoch: 58 Training Loss: 3.495231 Validation Loss: 3.219593 Training Accuracy: 0.952173 Validation Accuracy: 0.960952
Epoch: 59 Training Loss: 3.549008 Validation Loss: 2.765509 Training Accuracy: 0.956905 Validation Accuracy: 0.960659
Epoch: 60 Training Loss: 3.687503 Validation Loss: 2.706749 Training Accuracy: 0.950506 Validation Accuracy: 0.970476
Epoch: 61 Training Loss: 3.541313 Validation Loss: 3.154986 Training Accuracy: 0.952858 Validation Accuracy: 0.956996
Epoch: 62 Training Loss: 3.506211 Validation Loss: 2.924364 Training Accuracy: 0.956191 Validation Accuracy: 0.961612
Epoch: 63 Training Loss: 3.466941 Validation Loss: 2.835941 Training Accuracy: 0.954613 Validation Accuracy: 0.960952
Epoch: 64 Training Loss: 3.571767 Validation Loss: 3.051790 Training Accuracy: 0.950030 Validation Accuracy: 0.962857
Epoch: 65 Training Loss: 3.328473 Validation Loss: 2.663380 Training Accuracy: 0.961220 Validation Accuracy: 0.971282
Epoch: 66 Training Loss: 3.475173 Validation Loss: 2.737613 Training Accuracy: 0.958601 Validation Accuracy: 0.966520
Epoch: 67 Training Loss: 3.583273 Validation Loss: 2.931132 Training Accuracy: 0.946429 Validation Accuracy: 0.969377
Epoch: 68 Training Loss: 3.322832 Validation Loss: 3.299993 Training Accuracy: 0.955715 Validation Accuracy: 0.957656
Epoch: 69 Training Loss: 3.537165 Validation Loss: 2.970462 Training Accuracy: 0.959048 Validation Accuracy: 0.960952
Epoch: 70 Training Loss: 3.736241 Validation Loss: 2.907291 Training Accuracy: 0.949405 Validation Accuracy: 0.968571
Epoch: 71 Training Loss: 3.667625 Validation Loss: 2.714283 Training Accuracy: 0.953334 Validation Accuracy: 0.961612
Epoch: 72 Training Loss: 3.701843 Validation Loss: 2.829599 Training Accuracy: 0.953334 Validation Accuracy: 0.966374
Epoch: 73 Training Loss: 3.553051 Validation Loss: 3.216197 Training Accuracy: 0.954792 Validation Accuracy: 0.960806
Epoch: 74 Training Loss: 3.525238 Validation Loss: 2.870284 Training Accuracy: 0.955030 Validation Accuracy: 0.965714
Epoch: 75 Training Loss: 3.577650 Validation Loss: 3.150309 Training Accuracy: 0.952649 Validation Accuracy: 0.959048
Epoch: 76 Training Loss: 3.493605 Validation Loss: 3.120373 Training Accuracy: 0.955715 Validation Accuracy: 0.959048
Epoch: 77 Training Loss: 3.480341 Validation Loss: 2.993176 Training Accuracy: 0.952887 Validation Accuracy: 0.964762
Epoch: 78 Training Loss: 3.390446 Validation Loss: 3.006025 Training Accuracy: 0.955477 Validation Accuracy: 0.963663
Epoch: 79 Training Loss: 3.628780 Validation Loss: 2.964340 Training Accuracy: 0.952649 Validation Accuracy: 0.969084
Epoch: 80 Training Loss: 3.648959 Validation Loss: 3.007646 Training Accuracy: 0.951935 Validation Accuracy: 0.963810
Epoch: 81 Training Loss: 3.561923 Validation Loss: 2.625630 Training Accuracy: 0.950506 Validation Accuracy: 0.967619
Epoch: 82 Training Loss: 3.584743 Validation Loss: 3.547489 Training Accuracy: 0.956726 Validation Accuracy: 0.952234
Epoch: 83 Training Loss: 3.863186 Validation Loss: 2.812009 Training Accuracy: 0.950000 Validation Accuracy: 0.961905
Epoch: 84 Training Loss: 3.604072 Validation Loss: 2.893864 Training Accuracy: 0.951727 Validation Accuracy: 0.966520
Epoch: 85 Training Loss: 3.436723 Validation Loss: 3.245769 Training Accuracy: 0.954286 Validation Accuracy: 0.959707
Epoch: 86 Training Loss: 3.621120 Validation Loss: 2.688342 Training Accuracy: 0.953840 Validation Accuracy: 0.971429
Epoch: 87 Training Loss: 3.588997 Validation Loss: 2.520786 Training Accuracy: 0.949524 Validation Accuracy: 0.974286
Epoch: 88 Training Loss: 3.404798 Validation Loss: 2.762371 Training Accuracy: 0.956221 Validation Accuracy: 0.965421
Epoch: 89 Training Loss: 3.702103 Validation Loss: 3.059718 Training Accuracy: 0.950744 Validation Accuracy: 0.968571
Epoch: 90 Training Loss: 3.544047 Validation Loss: 2.799781 Training Accuracy: 0.955744 Validation Accuracy: 0.963809
Epoch: 91 Training Loss: 3.585321 Validation Loss: 3.098808 Training Accuracy: 0.952887 Validation Accuracy: 0.961612
Epoch: 92 Training Loss: 3.498460 Validation Loss: 2.919285 Training Accuracy: 0.953393 Validation Accuracy: 0.965714
Epoch: 93 Training Loss: 3.487115 Validation Loss: 2.431767 Training Accuracy: 0.954524 Validation Accuracy: 0.972088
Epoch: 94 Training Loss: 3.473630 Validation Loss: 3.140299 Training Accuracy: 0.953572 Validation Accuracy: 0.959048
Epoch: 95 Training Loss: 3.719396 Validation Loss: 2.712240 Training Accuracy: 0.952471 Validation Accuracy: 0.960806
Epoch: 96 Training Loss: 3.399465 Validation Loss: 2.774692 Training Accuracy: 0.956220 Validation Accuracy: 0.968278
Epoch: 97 Training Loss: 3.658062 Validation Loss: 2.984824 Training Accuracy: 0.951191 Validation Accuracy: 0.964762
Epoch: 98 Training Loss: 3.633016 Validation Loss: 3.182649 Training Accuracy: 0.955268 Validation Accuracy: 0.963663
Epoch: 99 Training Loss: 3.854623 Validation Loss: 2.827338 Training Accuracy: 0.949762 Validation Accuracy: 0.963809
Epoch: 100 Training Loss: 3.658060 Validation Loss: 3.395272 Training Accuracy: 0.953155 Validation Accuracy: 0.951428
plt.figure(figsize=(10,8))
plt.plot(train_losses,'--',linewidth = 2.5, label='Training loss')
plt.plot(valid_losses,'--', linewidth = 2.5, label='Validation loss')
plt.xlabel('Epochs',fontsize = 20)
plt.legend(fontsize = 14)
plt.xticks(fontsize = 12)
plt.yticks(fontsize = 12);
plt.figure(figsize=(10,8))
plt.plot(train_acc,'--', linewidth = 2.5, label='Training accuracy')
plt.plot(valid_acc,'--', linewidth = 2.5, label='Validation accuracy')
plt.xlabel('Epochs', fontsize = 20)
plt.legend(fontsize = 14)
plt.xticks(fontsize = 12)
plt.yticks(fontsize = 12);
res_model.load_state_dict(torch.load('./../models/res_model_pneumonia.pt'))
<All keys matched successfully>
# track test loss
test_loss = 0.0
class_correct = list(0. for i in range(2))
class_total = list(0. for i in range(2))
res_model.eval()
# iterating over test data
for data, target in tqdm(test_loader, desc = 'Model testing: '):
# moving tensors to GPU if CUDA is available
if train_on_gpu:
data, target = data.cuda(), target.cuda()
# forward pass: computing predicted outputs by passing inputs to the model
output = res_model(data)
target = target.long()
# calculating the batch loss
loss = criterion(output, target)
# updating test loss
test_loss += loss.item()*data.size(0)
# converting output probabilities to predicted class
_, pred = torch.max(output, 1)
# comparing predictions to true label
correct_tensor = pred.eq(target.data.view_as(pred))
correct = np.squeeze(correct_tensor.numpy()) if not train_on_gpu else np.squeeze(correct_tensor.cpu().numpy())
# calculating test accuracy for each object class
for i in range(len(target.data)):
label = target.data[i]
class_correct[label] += correct[i].item()
class_total[label] += 1
# average test loss
test_loss = test_loss/len(test_loader.dataset)
print('Test Loss: {:.6f}\n'.format(test_loss))
for i in range(2):
if class_total[i] > 0:
print('Test Accuracy of %5s: %2d%% (%2d/%2d)' % (
classes[i], 100 * class_correct[i] / class_total[i],
np.sum(class_correct[i]), np.sum(class_total[i])))
else:
print('Test Accuracy of %5s: N/A (no training examples)' % (classes[i]))
print('\nTest Accuracy (Overall): %2d%% (%2d/%2d)' % (
100. * np.sum(class_correct) / np.sum(class_total),
np.sum(class_correct), np.sum(class_total)))
Test Loss: 0.291028 Test Accuracy of PNEUMONIA: 77% (182/234) Test Accuracy of NORMAL: 96% (377/390) Test Accuracy (Overall): 89% (559/624)
confusion_matrix = torch.zeros(2,2)
with torch.no_grad():
for i, (data, target) in enumerate(test_loader):
if train_on_gpu:
data, target = data.cuda(), target.cuda()
output = res_model(data)
target = target.long()
_, preds = torch.max(output, 1)
for t, p in zip(target.view(-1), preds.view(-1)):
confusion_matrix[t.long(), p.long()] += 1
print(confusion_matrix)
tensor([[ 6., 1.],
[ 0., 17.]])
cm = confusion_matrix.numpy()
tn, fp, fn, tp = cm.ravel()
precision = tp/(tp+fp)
recall = tp/(tp+fn)
f1_score = (2*precision*recall)/(precision+recall)
print("Precision: {:.4f}\nRecall: {:.4f}\nF1 Score: {:.4f}".format(precision,recall,f1_score))
Precision: 0.9444 Recall: 1.0000 F1 Score: 0.9714
plt.figure(figsize=(8,7))
xticks = np.array(['NORMAL','PNEUMONIA'])
sns.heatmap(cm, annot = True, xticklabels = xticks, yticklabels = xticks)
plt.xlabel('PREDICTED',fontsize = 15)
plt.ylabel('TRUE',fontsize = 15);
# obtaining one batch of test images
data, target = next(iter(test_loader))
data.numpy()
# moving model inputs to cuda, if GPU available
if train_on_gpu:
data = data.cuda()
# getting sample outputs
output = res_model(data)
# converting output probabilities to predicted class
_, preds_tensor = torch.max(output, 1)
preds = np.squeeze(preds_tensor.numpy()) if not train_on_gpu else np.squeeze(preds_tensor.cpu().numpy())
# plotting the images in the batch, along with predicted and true labels
fig = plt.figure(figsize=(20,10))
for idx in np.arange(8):
ax = fig.add_subplot(2,4, idx+1, xticks=[], yticks=[])
ax.imshow(data[idx].permute(1, 2, 0) if not train_on_gpu else data[idx].cpu().permute(1, 2, 0))
ax.set_title("{} ({})".format(classes[preds[idx]], classes[target[idx]]),
color=("green" if preds[idx]==target[idx] else "red"))
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
vgg16 = models.vgg16(pretrained=True)
print(vgg16)
VGG(
(features): Sequential(
(0): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): ReLU(inplace=True)
(2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(3): ReLU(inplace=True)
(4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(5): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(6): ReLU(inplace=True)
(7): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(8): ReLU(inplace=True)
(9): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(10): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(11): ReLU(inplace=True)
(12): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(13): ReLU(inplace=True)
(14): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(15): ReLU(inplace=True)
(16): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(17): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(18): ReLU(inplace=True)
(19): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(20): ReLU(inplace=True)
(21): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(22): ReLU(inplace=True)
(23): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(24): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(25): ReLU(inplace=True)
(26): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(27): ReLU(inplace=True)
(28): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(29): ReLU(inplace=True)
(30): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
)
(avgpool): AdaptiveAvgPool2d(output_size=(7, 7))
(classifier): Sequential(
(0): Linear(in_features=25088, out_features=4096, bias=True)
(1): ReLU(inplace=True)
(2): Dropout(p=0.5, inplace=False)
(3): Linear(in_features=4096, out_features=4096, bias=True)
(4): ReLU(inplace=True)
(5): Dropout(p=0.5, inplace=False)
(6): Linear(in_features=4096, out_features=1000, bias=True)
)
)
# Freezing training for all "features" layers
for param in vgg16.features.parameters():
param.requires_grad = False
# Adding final layers
vgg16.fc = nn.Sequential(nn.Linear(4096,512),
nn.BatchNorm1d(512),
nn.ReLU(),
nn.Dropout(0.5),
nn.Linear(512,2),
nn.LogSoftmax(dim=1))
# specifying the loss function (categorical cross-entropy)
criterion = nn.CrossEntropyLoss()
# specifying optimizer
optimizer = optim.SGD(vgg16.parameters(), lr = 0.008,momentum = 0.9, nesterov = True)
# specifying scheduler
scheduler = optim.lr_scheduler.ExponentialLR(optimizer,gamma = 0.2)
if train_on_gpu:
vgg16.cuda()
print(vgg16)
VGG(
(features): Sequential(
(0): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): ReLU(inplace=True)
(2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(3): ReLU(inplace=True)
(4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(5): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(6): ReLU(inplace=True)
(7): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(8): ReLU(inplace=True)
(9): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(10): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(11): ReLU(inplace=True)
(12): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(13): ReLU(inplace=True)
(14): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(15): ReLU(inplace=True)
(16): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(17): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(18): ReLU(inplace=True)
(19): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(20): ReLU(inplace=True)
(21): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(22): ReLU(inplace=True)
(23): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(24): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(25): ReLU(inplace=True)
(26): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(27): ReLU(inplace=True)
(28): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(29): ReLU(inplace=True)
(30): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
)
(avgpool): AdaptiveAvgPool2d(output_size=(7, 7))
(classifier): Sequential(
(0): Linear(in_features=25088, out_features=4096, bias=True)
(1): ReLU(inplace=True)
(2): Dropout(p=0.5, inplace=False)
(3): Linear(in_features=4096, out_features=4096, bias=True)
(4): ReLU(inplace=True)
(5): Dropout(p=0.5, inplace=False)
(6): Linear(in_features=4096, out_features=1000, bias=True)
)
(fc): Sequential(
(0): Linear(in_features=4096, out_features=512, bias=True)
(1): BatchNorm1d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
(3): Dropout(p=0.5, inplace=False)
(4): Linear(in_features=512, out_features=2, bias=True)
(5): LogSoftmax(dim=1)
)
)
# number of epochs to train the model
n_epochs = 100
valid_loss_min = np.Inf # track change in validation loss
train_losses, valid_losses = [], []
train_acc, valid_acc = [], []
for epoch in tqdm(range(1, n_epochs+1), desc = 'Model Training/Epoch: '):
# keeping track of loss and accuracy
train_loss = 0.0
valid_loss = 0.0
train_accuracy = 0.0
valid_accuracy = 0.0
######################
# training the model #
######################
vgg16.train()
for data, target in tqdm(train_loader, desc = 'Model train: '):
# moving tensors to GPU if CUDA is available
if train_on_gpu:
data, target = data.cuda(), target.cuda()
# clearing the gradients of all optimized variables
optimizer.zero_grad()
# forward pass: computing predicted outputs by passing inputs to the model
output = vgg16(data)
target = target.long()
# calculating the batch loss
loss = criterion(output, target)
# backward pass: computing gradient of the loss with respect to model parameters
loss.backward()
# performing a single optimization step (parameter update)
optimizer.step()
# updating training loss
train_loss += loss.item()*data.size(0)
# updating training accuracy
train_accuracy += accuracy(output,target)
scheduler.step() # decay LR
########################
# validating the model #
########################
vgg16.eval()
for data, target in tqdm(valid_loader, desc = 'Model validation: '):
# moving tensors to GPU if CUDA is available
if train_on_gpu:
data, target = data.cuda(), target.cuda()
# forward pass: computing predicted outputs by passing inputs to the model
output = vgg16(data)
target = target.long()
# calculating the batch loss
loss = criterion(output, target)
# updating average validation loss
valid_loss += loss.item()*data.size(0)
# updating validation accuracy
valid_accuracy += accuracy(output,target)
# calculating average losses
train_loss = train_loss/len(train_loader)
valid_loss = valid_loss/len(valid_loader)
# calculating average accuracy
train_accuracy = train_accuracy/len(train_loader)
valid_accuracy = valid_accuracy/len(valid_loader)
train_losses.append(train_loss)
valid_losses.append(valid_loss)
train_acc.append(train_accuracy)
valid_acc.append(valid_accuracy)
# printing training/validation statistics
print('Epoch: {} \tTraining Loss: {:.6f} \tValidation Loss: {:.6f} \n\t\tTraining Accuracy: {:.6f} \tValidation Accuracy: {:.6f}'.format(
epoch, train_loss, valid_loss, train_accuracy, valid_accuracy))
# saving model if validation loss has decreased
if valid_loss <= valid_loss_min:
print('Validation loss decreased ({:.6f} --> {:.6f}). Saving model ...'.format(
valid_loss_min,
valid_loss))
torch.save(vgg16.state_dict(), './../models/vgg16_pneumonia.pt')
valid_loss_min = valid_loss
Epoch: 1 Training Loss: 17.667350 Validation Loss: 3.737621 Training Accuracy: 0.877858 Validation Accuracy: 0.947326 Validation loss decreased (inf --> 3.737621). Saving model ...
Epoch: 2 Training Loss: 4.178199 Validation Loss: 2.551276 Training Accuracy: 0.949614 Validation Accuracy: 0.969377 Validation loss decreased (3.737621 --> 2.551276). Saving model ...
Epoch: 3 Training Loss: 3.317313 Validation Loss: 2.644777 Training Accuracy: 0.957917 Validation Accuracy: 0.967473
Epoch: 4 Training Loss: 3.085204 Validation Loss: 2.578371 Training Accuracy: 0.963125 Validation Accuracy: 0.969377
Epoch: 5 Training Loss: 3.165493 Validation Loss: 2.704832 Training Accuracy: 0.960953 Validation Accuracy: 0.970330
Epoch: 6 Training Loss: 3.229269 Validation Loss: 2.836529 Training Accuracy: 0.963572 Validation Accuracy: 0.962857
Epoch: 7 Training Loss: 3.059022 Validation Loss: 2.540645 Training Accuracy: 0.962411 Validation Accuracy: 0.965568 Validation loss decreased (2.551276 --> 2.540645). Saving model ...
Epoch: 8 Training Loss: 3.013145 Validation Loss: 2.771429 Training Accuracy: 0.962917 Validation Accuracy: 0.961758
Epoch: 9 Training Loss: 3.195320 Validation Loss: 2.713270 Training Accuracy: 0.961220 Validation Accuracy: 0.970330
Epoch: 10 Training Loss: 2.864906 Validation Loss: 2.640942 Training Accuracy: 0.965060 Validation Accuracy: 0.966520
Epoch: 11 Training Loss: 3.058229 Validation Loss: 2.645853 Training Accuracy: 0.962887 Validation Accuracy: 0.963663
Epoch: 12 Training Loss: 3.188041 Validation Loss: 2.580250 Training Accuracy: 0.959107 Validation Accuracy: 0.969377
Epoch: 13 Training Loss: 3.293026 Validation Loss: 2.529081 Training Accuracy: 0.962173 Validation Accuracy: 0.968425 Validation loss decreased (2.540645 --> 2.529081). Saving model ...
Epoch: 14 Training Loss: 3.075837 Validation Loss: 2.707670 Training Accuracy: 0.962679 Validation Accuracy: 0.965421
Epoch: 15 Training Loss: 3.089580 Validation Loss: 2.626421 Training Accuracy: 0.960476 Validation Accuracy: 0.972088
Epoch: 16 Training Loss: 3.201968 Validation Loss: 2.437015 Training Accuracy: 0.960715 Validation Accuracy: 0.968571 Validation loss decreased (2.529081 --> 2.437015). Saving model ...
Epoch: 17 Training Loss: 3.028969 Validation Loss: 2.686520 Training Accuracy: 0.962381 Validation Accuracy: 0.967619
Epoch: 18 Training Loss: 3.078695 Validation Loss: 2.777402 Training Accuracy: 0.960000 Validation Accuracy: 0.965421
Epoch: 19 Training Loss: 3.057311 Validation Loss: 2.530627 Training Accuracy: 0.965030 Validation Accuracy: 0.970476
Epoch: 20 Training Loss: 2.879777 Validation Loss: 2.599185 Training Accuracy: 0.964792 Validation Accuracy: 0.968278
Epoch: 21 Training Loss: 3.171468 Validation Loss: 2.506772 Training Accuracy: 0.960238 Validation Accuracy: 0.968278
Epoch: 22 Training Loss: 3.144134 Validation Loss: 2.307255 Training Accuracy: 0.962619 Validation Accuracy: 0.969377 Validation loss decreased (2.437015 --> 2.307255). Saving model ...
Epoch: 23 Training Loss: 3.090208 Validation Loss: 2.330842 Training Accuracy: 0.962917 Validation Accuracy: 0.968571
Epoch: 24 Training Loss: 3.272624 Validation Loss: 2.495125 Training Accuracy: 0.959346 Validation Accuracy: 0.969377
Epoch: 25 Training Loss: 3.465077 Validation Loss: 2.216696 Training Accuracy: 0.954375 Validation Accuracy: 0.971429 Validation loss decreased (2.307255 --> 2.216696). Saving model ...
Epoch: 26 Training Loss: 3.084881 Validation Loss: 2.701689 Training Accuracy: 0.957143 Validation Accuracy: 0.963809
Epoch: 27 Training Loss: 3.038011 Validation Loss: 2.643992 Training Accuracy: 0.960953 Validation Accuracy: 0.968425
Epoch: 28 Training Loss: 3.050145 Validation Loss: 2.572652 Training Accuracy: 0.960744 Validation Accuracy: 0.969524
Epoch: 29 Training Loss: 3.352974 Validation Loss: 2.464735 Training Accuracy: 0.958393 Validation Accuracy: 0.971429
Epoch: 30 Training Loss: 3.131793 Validation Loss: 2.631704 Training Accuracy: 0.963334 Validation Accuracy: 0.964762
Epoch: 31 Training Loss: 3.210741 Validation Loss: 2.712665 Training Accuracy: 0.960000 Validation Accuracy: 0.969377
Epoch: 32 Training Loss: 3.097784 Validation Loss: 2.453748 Training Accuracy: 0.960030 Validation Accuracy: 0.968571
Epoch: 33 Training Loss: 2.993579 Validation Loss: 2.700142 Training Accuracy: 0.960000 Validation Accuracy: 0.964615
Epoch: 34 Training Loss: 3.008549 Validation Loss: 2.708023 Training Accuracy: 0.961459 Validation Accuracy: 0.964616
Epoch: 35 Training Loss: 3.224409 Validation Loss: 2.819792 Training Accuracy: 0.958661 Validation Accuracy: 0.967619
Epoch: 36 Training Loss: 3.206897 Validation Loss: 2.437948 Training Accuracy: 0.960477 Validation Accuracy: 0.969524
Epoch: 37 Training Loss: 2.940715 Validation Loss: 2.613734 Training Accuracy: 0.961935 Validation Accuracy: 0.967472
Epoch: 38 Training Loss: 2.964604 Validation Loss: 2.667775 Training Accuracy: 0.963810 Validation Accuracy: 0.966520
Epoch: 39 Training Loss: 3.066457 Validation Loss: 2.635818 Training Accuracy: 0.963571 Validation Accuracy: 0.962564
Epoch: 40 Training Loss: 3.274643 Validation Loss: 2.293757 Training Accuracy: 0.961221 Validation Accuracy: 0.966667
Epoch: 41 Training Loss: 3.145790 Validation Loss: 2.705035 Training Accuracy: 0.962649 Validation Accuracy: 0.966374
Epoch: 42 Training Loss: 3.185913 Validation Loss: 2.724364 Training Accuracy: 0.963095 Validation Accuracy: 0.965714
Epoch: 43 Training Loss: 3.124463 Validation Loss: 2.504432 Training Accuracy: 0.960715 Validation Accuracy: 0.970476
Epoch: 44 Training Loss: 3.029422 Validation Loss: 2.544516 Training Accuracy: 0.961221 Validation Accuracy: 0.964469
Epoch: 45 Training Loss: 3.240280 Validation Loss: 2.597324 Training Accuracy: 0.958602 Validation Accuracy: 0.964615
Epoch: 46 Training Loss: 3.115454 Validation Loss: 2.316215 Training Accuracy: 0.961905 Validation Accuracy: 0.980000
Epoch: 47 Training Loss: 2.816770 Validation Loss: 2.352490 Training Accuracy: 0.964584 Validation Accuracy: 0.969377
Epoch: 48 Training Loss: 3.372668 Validation Loss: 2.554384 Training Accuracy: 0.955953 Validation Accuracy: 0.963516
Epoch: 49 Training Loss: 3.070006 Validation Loss: 2.557437 Training Accuracy: 0.962649 Validation Accuracy: 0.965421
Epoch: 50 Training Loss: 3.329403 Validation Loss: 2.594191 Training Accuracy: 0.959345 Validation Accuracy: 0.967619
Epoch: 51 Training Loss: 3.188780 Validation Loss: 2.588759 Training Accuracy: 0.961191 Validation Accuracy: 0.968425
Epoch: 52 Training Loss: 3.071174 Validation Loss: 2.539120 Training Accuracy: 0.960238 Validation Accuracy: 0.964469
Epoch: 53 Training Loss: 3.312355 Validation Loss: 2.842193 Training Accuracy: 0.958601 Validation Accuracy: 0.964616
Epoch: 54 Training Loss: 3.002052 Validation Loss: 2.548175 Training Accuracy: 0.963601 Validation Accuracy: 0.968425
Epoch: 55 Training Loss: 3.342660 Validation Loss: 2.738044 Training Accuracy: 0.961905 Validation Accuracy: 0.969524
Epoch: 56 Training Loss: 3.329132 Validation Loss: 2.625427 Training Accuracy: 0.961667 Validation Accuracy: 0.971282
Epoch: 57 Training Loss: 2.978728 Validation Loss: 2.803549 Training Accuracy: 0.961220 Validation Accuracy: 0.962711
Epoch: 58 Training Loss: 3.086485 Validation Loss: 2.710947 Training Accuracy: 0.960715 Validation Accuracy: 0.964762
Epoch: 59 Training Loss: 2.952445 Validation Loss: 2.523361 Training Accuracy: 0.961905 Validation Accuracy: 0.971282
Epoch: 60 Training Loss: 3.047914 Validation Loss: 2.708597 Training Accuracy: 0.962411 Validation Accuracy: 0.970476
Epoch: 61 Training Loss: 3.132514 Validation Loss: 2.607742 Training Accuracy: 0.963363 Validation Accuracy: 0.967180
Epoch: 62 Training Loss: 3.107331 Validation Loss: 2.515906 Training Accuracy: 0.960000 Validation Accuracy: 0.968425
Epoch: 63 Training Loss: 3.232370 Validation Loss: 2.358983 Training Accuracy: 0.958571 Validation Accuracy: 0.974945
Epoch: 64 Training Loss: 3.106707 Validation Loss: 2.614070 Training Accuracy: 0.957887 Validation Accuracy: 0.972234
Epoch: 65 Training Loss: 3.175229 Validation Loss: 2.786850 Training Accuracy: 0.960060 Validation Accuracy: 0.964762
Epoch: 66 Training Loss: 3.235923 Validation Loss: 2.581825 Training Accuracy: 0.958155 Validation Accuracy: 0.963809
Epoch: 67 Training Loss: 3.054971 Validation Loss: 2.880507 Training Accuracy: 0.960744 Validation Accuracy: 0.960806
Epoch: 68 Training Loss: 3.241127 Validation Loss: 2.349994 Training Accuracy: 0.960238 Validation Accuracy: 0.969231
Epoch: 69 Training Loss: 3.240683 Validation Loss: 2.695304 Training Accuracy: 0.959524 Validation Accuracy: 0.964616
Epoch: 70 Training Loss: 3.350330 Validation Loss: 2.726052 Training Accuracy: 0.959554 Validation Accuracy: 0.969524
Epoch: 71 Training Loss: 3.130748 Validation Loss: 2.859330 Training Accuracy: 0.964524 Validation Accuracy: 0.967473
Epoch: 72 Training Loss: 3.063134 Validation Loss: 2.639600 Training Accuracy: 0.962203 Validation Accuracy: 0.962710
Epoch: 73 Training Loss: 3.239754 Validation Loss: 2.465410 Training Accuracy: 0.959078 Validation Accuracy: 0.973993
Epoch: 74 Training Loss: 3.128110 Validation Loss: 2.636752 Training Accuracy: 0.960030 Validation Accuracy: 0.966374
Epoch: 75 Training Loss: 3.264820 Validation Loss: 2.534028 Training Accuracy: 0.960298 Validation Accuracy: 0.969377
Epoch: 76 Training Loss: 3.103342 Validation Loss: 2.579977 Training Accuracy: 0.961429 Validation Accuracy: 0.967619
Epoch: 77 Training Loss: 3.168887 Validation Loss: 2.510801 Training Accuracy: 0.959554 Validation Accuracy: 0.972234
Epoch: 78 Training Loss: 3.279042 Validation Loss: 2.703624 Training Accuracy: 0.957887 Validation Accuracy: 0.968278
Epoch: 79 Training Loss: 2.880642 Validation Loss: 2.765621 Training Accuracy: 0.963601 Validation Accuracy: 0.964762
Epoch: 80 Training Loss: 3.227795 Validation Loss: 2.657454 Training Accuracy: 0.960477 Validation Accuracy: 0.964762
Epoch: 81 Training Loss: 3.234135 Validation Loss: 2.479020 Training Accuracy: 0.960715 Validation Accuracy: 0.965421
Epoch: 82 Training Loss: 3.084176 Validation Loss: 2.702847 Training Accuracy: 0.963096 Validation Accuracy: 0.969231
Epoch: 83 Training Loss: 2.980287 Validation Loss: 2.154894 Training Accuracy: 0.961429 Validation Accuracy: 0.976190 Validation loss decreased (2.216696 --> 2.154894). Saving model ...
Epoch: 84 Training Loss: 3.100980 Validation Loss: 2.506936 Training Accuracy: 0.960000 Validation Accuracy: 0.968425
Epoch: 85 Training Loss: 3.017507 Validation Loss: 2.597001 Training Accuracy: 0.963661 Validation Accuracy: 0.972234
Epoch: 86 Training Loss: 2.962877 Validation Loss: 2.433483 Training Accuracy: 0.962857 Validation Accuracy: 0.968571
Epoch: 87 Training Loss: 3.190441 Validation Loss: 2.720800 Training Accuracy: 0.959524 Validation Accuracy: 0.968571
Epoch: 88 Training Loss: 3.147300 Validation Loss: 2.570401 Training Accuracy: 0.959286 Validation Accuracy: 0.967619
Epoch: 89 Training Loss: 3.232334 Validation Loss: 2.645331 Training Accuracy: 0.958810 Validation Accuracy: 0.971282
Epoch: 90 Training Loss: 3.267955 Validation Loss: 2.681733 Training Accuracy: 0.961250 Validation Accuracy: 0.969524
Epoch: 91 Training Loss: 3.189661 Validation Loss: 2.560663 Training Accuracy: 0.960268 Validation Accuracy: 0.971429
Epoch: 92 Training Loss: 3.080562 Validation Loss: 2.638019 Training Accuracy: 0.961905 Validation Accuracy: 0.965421
Epoch: 93 Training Loss: 3.319650 Validation Loss: 2.569783 Training Accuracy: 0.961697 Validation Accuracy: 0.969377
Epoch: 94 Training Loss: 3.181800 Validation Loss: 2.748830 Training Accuracy: 0.961667 Validation Accuracy: 0.962857
Epoch: 95 Training Loss: 3.117088 Validation Loss: 2.466863 Training Accuracy: 0.957679 Validation Accuracy: 0.967473
Epoch: 96 Training Loss: 3.198888 Validation Loss: 2.496233 Training Accuracy: 0.959078 Validation Accuracy: 0.968425
Epoch: 97 Training Loss: 3.200422 Validation Loss: 2.530464 Training Accuracy: 0.956905 Validation Accuracy: 0.967619
Epoch: 98 Training Loss: 3.093969 Validation Loss: 2.485638 Training Accuracy: 0.959524 Validation Accuracy: 0.970476
Epoch: 99 Training Loss: 3.180388 Validation Loss: 2.550592 Training Accuracy: 0.957887 Validation Accuracy: 0.966374
Epoch: 100 Training Loss: 3.258148 Validation Loss: 2.847490 Training Accuracy: 0.959316 Validation Accuracy: 0.964762
plt.figure(figsize=(10,8))
plt.plot(train_losses,'--',linewidth = 2.5, label='Training loss')
plt.plot(valid_losses,'--', linewidth = 2.5, label='Validation loss')
plt.xlabel('Epochs',fontsize = 20)
plt.legend(fontsize = 14)
plt.xticks(fontsize = 12)
plt.yticks(fontsize = 12);
plt.figure(figsize=(10,8))
plt.plot(train_acc,'--', linewidth = 2.5, label='Training accuracy')
plt.plot(valid_acc,'--', linewidth = 2.5, label='Validation accuracy')
plt.xlabel('Epochs', fontsize = 20)
plt.legend(fontsize = 14)
plt.xticks(fontsize = 12)
plt.yticks(fontsize = 12);
vgg16.load_state_dict(torch.load('./../models/vgg16_pneumonia.pt'))
<All keys matched successfully>
# track test loss
test_loss = 0.0
class_correct = list(0. for i in range(2))
class_total = list(0. for i in range(2))
vgg16.eval()
# iterating over test data
for data, target in tqdm(test_loader, desc = 'Model testing: '):
# moving tensors to GPU if CUDA is available
if train_on_gpu:
data, target = data.cuda(), target.cuda()
# forward pass: computing predicted outputs by passing inputs to the model
output = vgg16(data)
target = target.long()
# calculating the batch loss
loss = criterion(output, target)
# updating test loss
test_loss += loss.item()*data.size(0)
# converting output probabilities to predicted class
_, pred = torch.max(output, 1)
# comparing predictions to true label
correct_tensor = pred.eq(target.data.view_as(pred))
correct = np.squeeze(correct_tensor.numpy()) if not train_on_gpu else np.squeeze(correct_tensor.cpu().numpy())
# calculating test accuracy for each object class
for i in range(len(target.data)):
label = target.data[i]
class_correct[label] += correct[i].item()
class_total[label] += 1
# average test loss
test_loss = test_loss/len(test_loader.dataset)
print('Test Loss: {:.6f}\n'.format(test_loss))
for i in range(2):
if class_total[i] > 0:
print('Test Accuracy of %5s: %2d%% (%2d/%2d)' % (
classes[i], 100 * class_correct[i] / class_total[i],
np.sum(class_correct[i]), np.sum(class_total[i])))
else:
print('Test Accuracy of %5s: N/A (no training examples)' % (classes[i]))
print('\nTest Accuracy (Overall): %2d%% (%2d/%2d)' % (
100. * np.sum(class_correct) / np.sum(class_total),
np.sum(class_correct), np.sum(class_total)))
Test Loss: 0.213057 Test Accuracy of PNEUMONIA: 81% (191/234) Test Accuracy of NORMAL: 96% (378/390) Test Accuracy (Overall): 91% (569/624)
confusion_matrix = torch.zeros(2,2)
with torch.no_grad():
for i, (data, target) in enumerate(test_loader):
if train_on_gpu:
data, target = data.cuda(), target.cuda()
output = vgg16(data)
target = target.long()
_, preds = torch.max(output, 1)
for t, p in zip(target.view(-1), preds.view(-1)):
confusion_matrix[t.long(), p.long()] += 1
print(confusion_matrix)
tensor([[191., 43.],
[ 12., 378.]])
cm = confusion_matrix.numpy()
tn, fp, fn, tp = cm.ravel()
precision = tp/(tp+fp)
recall = tp/(tp+fn)
f1_score = (2*precision*recall)/(precision+recall)
print("Precision: {:.4f}\nRecall: {:.4f}\nF1 Score: {:.4f}".format(precision,recall,f1_score))
Precision: 0.8979 Recall: 0.9692 F1 Score: 0.9322
plt.figure(figsize=(8,7))
xticks = np.array(['NORMAL','PNEUMONIA'])
sns.heatmap(cm, annot = True,
xticklabels = xticks, yticklabels = xticks)
plt.xlabel('PREDICTED',fontsize = 15)
plt.ylabel('TRUE',fontsize = 15);
Text(51.0, 0.5, 'TRUE')
# obtaining one batch of test images
data, target = next(iter(test_loader))
data.numpy()
# moving model inputs to cuda, if GPU available
if train_on_gpu:
data = data.cuda()
# getting sample outputs
output = vgg16(data)
# converting output probabilities to predicted class
_, preds_tensor = torch.max(output, 1)
preds = np.squeeze(preds_tensor.numpy()) if not train_on_gpu else np.squeeze(preds_tensor.cpu().numpy())
# plotting the images in the batch, along with predicted and true labels
fig = plt.figure(figsize=(20,10))
for idx in np.arange(8):
ax = fig.add_subplot(2,4, idx+1, xticks=[], yticks=[])
ax.imshow(data[idx].permute(1, 2, 0) if not train_on_gpu else data[idx].cpu().permute(1, 2, 0))
ax.set_title("{} ({})".format(classes[preds[idx]], classes[target[idx]]),
color=("green" if preds[idx]==target[idx] else "red"))
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
vgg19 = models.vgg19(pretrained=True)
print(vgg19)
Downloading: "https://download.pytorch.org/models/vgg19-dcbb9e9d.pth" to /home/sysadm/.cache/torch/hub/checkpoints/vgg19-dcbb9e9d.pth
VGG(
(features): Sequential(
(0): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): ReLU(inplace=True)
(2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(3): ReLU(inplace=True)
(4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(5): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(6): ReLU(inplace=True)
(7): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(8): ReLU(inplace=True)
(9): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(10): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(11): ReLU(inplace=True)
(12): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(13): ReLU(inplace=True)
(14): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(15): ReLU(inplace=True)
(16): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(17): ReLU(inplace=True)
(18): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(19): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(20): ReLU(inplace=True)
(21): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(22): ReLU(inplace=True)
(23): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(24): ReLU(inplace=True)
(25): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(26): ReLU(inplace=True)
(27): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(28): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(29): ReLU(inplace=True)
(30): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(31): ReLU(inplace=True)
(32): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(33): ReLU(inplace=True)
(34): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(35): ReLU(inplace=True)
(36): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
)
(avgpool): AdaptiveAvgPool2d(output_size=(7, 7))
(classifier): Sequential(
(0): Linear(in_features=25088, out_features=4096, bias=True)
(1): ReLU(inplace=True)
(2): Dropout(p=0.5, inplace=False)
(3): Linear(in_features=4096, out_features=4096, bias=True)
(4): ReLU(inplace=True)
(5): Dropout(p=0.5, inplace=False)
(6): Linear(in_features=4096, out_features=1000, bias=True)
)
)
# Freezing training for all "features" layers
for param in vgg19.features.parameters():
param.requires_grad = False
# Adding final layers
vgg19.fc = nn.Sequential(nn.Linear(4096,512),
nn.BatchNorm1d(512),
nn.ReLU(),
nn.Dropout(0.5),
nn.Linear(512,2),
nn.LogSoftmax(dim=1))
# specifying the loss function (categorical cross-entropy)
criterion = nn.CrossEntropyLoss()
# specifying optimizer
optimizer = optim.SGD(vgg19.parameters(), lr = 0.008, momentum = 0.9, nesterov = True)
# specifying scheduler
scheduler = optim.lr_scheduler.ExponentialLR(optimizer,gamma = 0.2)
if train_on_gpu:
vgg19.cuda()
print(vgg19)
VGG(
(features): Sequential(
(0): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(1): ReLU(inplace=True)
(2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(3): ReLU(inplace=True)
(4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(5): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(6): ReLU(inplace=True)
(7): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(8): ReLU(inplace=True)
(9): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(10): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(11): ReLU(inplace=True)
(12): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(13): ReLU(inplace=True)
(14): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(15): ReLU(inplace=True)
(16): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(17): ReLU(inplace=True)
(18): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(19): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(20): ReLU(inplace=True)
(21): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(22): ReLU(inplace=True)
(23): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(24): ReLU(inplace=True)
(25): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(26): ReLU(inplace=True)
(27): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
(28): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(29): ReLU(inplace=True)
(30): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(31): ReLU(inplace=True)
(32): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(33): ReLU(inplace=True)
(34): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
(35): ReLU(inplace=True)
(36): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
)
(avgpool): AdaptiveAvgPool2d(output_size=(7, 7))
(classifier): Sequential(
(0): Linear(in_features=25088, out_features=4096, bias=True)
(1): ReLU(inplace=True)
(2): Dropout(p=0.5, inplace=False)
(3): Linear(in_features=4096, out_features=4096, bias=True)
(4): ReLU(inplace=True)
(5): Dropout(p=0.5, inplace=False)
(6): Linear(in_features=4096, out_features=1000, bias=True)
)
(fc): Sequential(
(0): Linear(in_features=4096, out_features=512, bias=True)
(1): BatchNorm1d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
(2): ReLU()
(3): Dropout(p=0.5, inplace=False)
(4): Linear(in_features=512, out_features=2, bias=True)
(5): LogSoftmax(dim=1)
)
)
# number of epochs to train the model
n_epochs = 100
valid_loss_min = np.Inf # track change in validation loss
train_losses, valid_losses = [], []
train_acc, valid_acc = [], []
for epoch in tqdm(range(1, n_epochs+1), desc = 'Model Training/Epoch: '):
# keeping track of loss and accuracy
train_loss = 0.0
valid_loss = 0.0
train_accuracy = 0.0
valid_accuracy = 0.0
######################
# training the model #
######################
vgg19.train()
for data, target in tqdm(train_loader, desc = 'Model train: '):
# moving tensors to GPU if CUDA is available
if train_on_gpu:
data, target = data.cuda(), target.cuda()
# clearing the gradients of all optimized variables
optimizer.zero_grad()
# forward pass: computing predicted outputs by passing inputs to the model
output = vgg19(data)
target = target.long()
# calculating the batch loss
loss = criterion(output, target)
# backward pass: computing gradient of the loss with respect to model parameters
loss.backward()
# performing a single optimization step (parameter update)
optimizer.step()
# updating training loss
train_loss += loss.item()*data.size(0)
# updating training accuracy
train_accuracy += accuracy(output,target)
scheduler.step() # decay LR
########################
# validating the model #
########################
vgg19.eval()
for data, target in tqdm(valid_loader, desc = 'Model validation: '):
# moving tensors to GPU if CUDA is available
if train_on_gpu:
data, target = data.cuda(), target.cuda()
# forward pass: computing predicted outputs by passing inputs to the model
output = vgg19(data)
target = target.long()
# calculating the batch loss
loss = criterion(output, target)
# updating average validation loss
valid_loss += loss.item()*data.size(0)
# updating validation accuracy
valid_accuracy += accuracy(output,target)
# calculating average losses
train_loss = train_loss/len(train_loader)
valid_loss = valid_loss/len(valid_loader)
# calculating average accuracy
train_accuracy = train_accuracy/len(train_loader)
valid_accuracy = valid_accuracy/len(valid_loader)
train_losses.append(train_loss)
valid_losses.append(valid_loss)
train_acc.append(train_accuracy)
valid_acc.append(valid_accuracy)
# printing training/validation statistics
print('Epoch: {} \tTraining Loss: {:.6f} \tValidation Loss: {:.6f} \n\t\tTraining Accuracy: {:.6f} \tValidation Accuracy: {:.6f}'.format(
epoch, train_loss, valid_loss, train_accuracy, valid_accuracy))
# saving model if validation loss has decreased
if valid_loss <= valid_loss_min:
print('Validation loss decreased ({:.6f} --> {:.6f}). Saving model ...'.format(
valid_loss_min,
valid_loss))
torch.save(vgg19.state_dict(), './../models/vgg19_pneumonia.pt')
valid_loss_min = valid_loss
Epoch: 1 Training Loss: 13.510141 Validation Loss: 3.935669 Training Accuracy: 0.882679 Validation Accuracy: 0.948425 Validation loss decreased (inf --> 3.935669). Saving model ...
Epoch: 2 Training Loss: 4.334541 Validation Loss: 3.558420 Training Accuracy: 0.949524 Validation Accuracy: 0.952234 Validation loss decreased (3.935669 --> 3.558420). Saving model ...
Epoch: 3 Training Loss: 3.454049 Validation Loss: 3.205781 Training Accuracy: 0.953363 Validation Accuracy: 0.958095 Validation loss decreased (3.558420 --> 3.205781). Saving model ...
Epoch: 4 Training Loss: 3.415587 Validation Loss: 2.847211 Training Accuracy: 0.954762 Validation Accuracy: 0.960806 Validation loss decreased (3.205781 --> 2.847211). Saving model ...
Epoch: 5 Training Loss: 3.489104 Validation Loss: 3.607472 Training Accuracy: 0.954048 Validation Accuracy: 0.951795
Epoch: 6 Training Loss: 3.380298 Validation Loss: 3.082811 Training Accuracy: 0.960476 Validation Accuracy: 0.956044
Epoch: 7 Training Loss: 3.605634 Validation Loss: 2.994798 Training Accuracy: 0.955744 Validation Accuracy: 0.961465
Epoch: 8 Training Loss: 3.310773 Validation Loss: 2.919432 Training Accuracy: 0.957887 Validation Accuracy: 0.956703
Epoch: 9 Training Loss: 3.245864 Validation Loss: 2.913105 Training Accuracy: 0.960060 Validation Accuracy: 0.962564
Epoch: 10 Training Loss: 3.459905 Validation Loss: 3.198183 Training Accuracy: 0.960715 Validation Accuracy: 0.958095
Epoch: 11 Training Loss: 3.613631 Validation Loss: 2.924762 Training Accuracy: 0.959286 Validation Accuracy: 0.960000
Epoch: 12 Training Loss: 3.612762 Validation Loss: 3.187192 Training Accuracy: 0.954316 Validation Accuracy: 0.954286
Epoch: 13 Training Loss: 3.284087 Validation Loss: 2.847460 Training Accuracy: 0.956250 Validation Accuracy: 0.962711
Epoch: 14 Training Loss: 3.242849 Validation Loss: 3.006502 Training Accuracy: 0.960268 Validation Accuracy: 0.962564
Epoch: 15 Training Loss: 3.386828 Validation Loss: 3.158714 Training Accuracy: 0.958125 Validation Accuracy: 0.952234
Epoch: 16 Training Loss: 3.312329 Validation Loss: 3.015409 Training Accuracy: 0.956697 Validation Accuracy: 0.959048
Epoch: 17 Training Loss: 3.560689 Validation Loss: 2.914869 Training Accuracy: 0.955268 Validation Accuracy: 0.961905
Epoch: 18 Training Loss: 3.533449 Validation Loss: 2.908084 Training Accuracy: 0.950715 Validation Accuracy: 0.958901
Epoch: 19 Training Loss: 3.471468 Validation Loss: 2.992906 Training Accuracy: 0.954554 Validation Accuracy: 0.956703
Epoch: 20 Training Loss: 3.430110 Validation Loss: 2.746312 Training Accuracy: 0.952619 Validation Accuracy: 0.958901 Validation loss decreased (2.847211 --> 2.746312). Saving model ...
Epoch: 21 Training Loss: 3.666111 Validation Loss: 3.123970 Training Accuracy: 0.951905 Validation Accuracy: 0.956996
Epoch: 22 Training Loss: 3.301889 Validation Loss: 2.963295 Training Accuracy: 0.955536 Validation Accuracy: 0.959048
Epoch: 23 Training Loss: 3.122262 Validation Loss: 3.019467 Training Accuracy: 0.958363 Validation Accuracy: 0.954286
Epoch: 24 Training Loss: 3.485370 Validation Loss: 2.955798 Training Accuracy: 0.955268 Validation Accuracy: 0.956996
Epoch: 25 Training Loss: 3.533084 Validation Loss: 3.317520 Training Accuracy: 0.955030 Validation Accuracy: 0.953187
Epoch: 26 Training Loss: 3.486697 Validation Loss: 2.964122 Training Accuracy: 0.956667 Validation Accuracy: 0.957143
Epoch: 27 Training Loss: 3.236516 Validation Loss: 3.061709 Training Accuracy: 0.959792 Validation Accuracy: 0.961905
Epoch: 28 Training Loss: 3.284637 Validation Loss: 3.184100 Training Accuracy: 0.958810 Validation Accuracy: 0.959048
Epoch: 29 Training Loss: 3.580308 Validation Loss: 3.108135 Training Accuracy: 0.954137 Validation Accuracy: 0.958755
Epoch: 30 Training Loss: 3.123207 Validation Loss: 3.051007 Training Accuracy: 0.957857 Validation Accuracy: 0.956997
Epoch: 31 Training Loss: 3.457726 Validation Loss: 2.941169 Training Accuracy: 0.954316 Validation Accuracy: 0.962857
Epoch: 32 Training Loss: 3.342324 Validation Loss: 2.993522 Training Accuracy: 0.958125 Validation Accuracy: 0.962564
Epoch: 33 Training Loss: 3.711888 Validation Loss: 2.902280 Training Accuracy: 0.954048 Validation Accuracy: 0.960952
Epoch: 34 Training Loss: 3.566735 Validation Loss: 2.902579 Training Accuracy: 0.956429 Validation Accuracy: 0.962857
Epoch: 35 Training Loss: 3.291976 Validation Loss: 3.068246 Training Accuracy: 0.956697 Validation Accuracy: 0.957143
Epoch: 36 Training Loss: 3.482515 Validation Loss: 2.886519 Training Accuracy: 0.956935 Validation Accuracy: 0.955897
Epoch: 37 Training Loss: 3.441705 Validation Loss: 3.070416 Training Accuracy: 0.958810 Validation Accuracy: 0.955897
Epoch: 38 Training Loss: 3.265039 Validation Loss: 2.918490 Training Accuracy: 0.960000 Validation Accuracy: 0.964762
Epoch: 39 Training Loss: 3.465814 Validation Loss: 2.951473 Training Accuracy: 0.957887 Validation Accuracy: 0.967619
Epoch: 40 Training Loss: 3.224375 Validation Loss: 3.218928 Training Accuracy: 0.956935 Validation Accuracy: 0.959853
Epoch: 41 Training Loss: 3.251793 Validation Loss: 3.368749 Training Accuracy: 0.958601 Validation Accuracy: 0.951941
Epoch: 42 Training Loss: 3.609659 Validation Loss: 3.005757 Training Accuracy: 0.950715 Validation Accuracy: 0.964615
Epoch: 43 Training Loss: 3.401568 Validation Loss: 2.871498 Training Accuracy: 0.957619 Validation Accuracy: 0.964322
Epoch: 44 Training Loss: 3.456453 Validation Loss: 3.193111 Training Accuracy: 0.959524 Validation Accuracy: 0.956996
Epoch: 45 Training Loss: 3.334325 Validation Loss: 3.158654 Training Accuracy: 0.953572 Validation Accuracy: 0.958901
Epoch: 46 Training Loss: 3.415161 Validation Loss: 2.848882 Training Accuracy: 0.959286 Validation Accuracy: 0.963663
Epoch: 47 Training Loss: 3.310765 Validation Loss: 2.942414 Training Accuracy: 0.958125 Validation Accuracy: 0.957949
Epoch: 48 Training Loss: 3.719572 Validation Loss: 3.049627 Training Accuracy: 0.948602 Validation Accuracy: 0.960806
Epoch: 49 Training Loss: 3.411166 Validation Loss: 2.855486 Training Accuracy: 0.955953 Validation Accuracy: 0.957509
Epoch: 50 Training Loss: 3.524796 Validation Loss: 2.829436 Training Accuracy: 0.955953 Validation Accuracy: 0.960000
Epoch: 51 Training Loss: 3.160231 Validation Loss: 2.928636 Training Accuracy: 0.961667 Validation Accuracy: 0.960952
Epoch: 52 Training Loss: 3.506489 Validation Loss: 3.113644 Training Accuracy: 0.955268 Validation Accuracy: 0.957949
Epoch: 53 Training Loss: 3.367503 Validation Loss: 3.250704 Training Accuracy: 0.956667 Validation Accuracy: 0.959048
Epoch: 54 Training Loss: 3.587550 Validation Loss: 2.905142 Training Accuracy: 0.955715 Validation Accuracy: 0.961758
Epoch: 55 Training Loss: 3.373432 Validation Loss: 2.996850 Training Accuracy: 0.956191 Validation Accuracy: 0.957949
Epoch: 56 Training Loss: 3.496185 Validation Loss: 2.971601 Training Accuracy: 0.955715 Validation Accuracy: 0.960659
Epoch: 57 Training Loss: 3.358374 Validation Loss: 3.140039 Training Accuracy: 0.955000 Validation Accuracy: 0.957143
Epoch: 58 Training Loss: 3.189660 Validation Loss: 3.249238 Training Accuracy: 0.959078 Validation Accuracy: 0.952381
Epoch: 59 Training Loss: 3.596932 Validation Loss: 2.730829 Training Accuracy: 0.952143 Validation Accuracy: 0.966520 Validation loss decreased (2.746312 --> 2.730829). Saving model ...
Epoch: 60 Training Loss: 3.436621 Validation Loss: 2.758686 Training Accuracy: 0.958810 Validation Accuracy: 0.967472
Epoch: 61 Training Loss: 3.511800 Validation Loss: 3.162646 Training Accuracy: 0.954762 Validation Accuracy: 0.955092
Epoch: 62 Training Loss: 3.441398 Validation Loss: 2.950811 Training Accuracy: 0.955715 Validation Accuracy: 0.961758
Epoch: 63 Training Loss: 3.377047 Validation Loss: 3.012627 Training Accuracy: 0.959614 Validation Accuracy: 0.965421
Epoch: 64 Training Loss: 3.310835 Validation Loss: 3.311706 Training Accuracy: 0.956429 Validation Accuracy: 0.952894
Epoch: 65 Training Loss: 3.330177 Validation Loss: 3.077992 Training Accuracy: 0.955506 Validation Accuracy: 0.951941
Epoch: 66 Training Loss: 3.330314 Validation Loss: 2.579198 Training Accuracy: 0.957649 Validation Accuracy: 0.970183 Validation loss decreased (2.730829 --> 2.579198). Saving model ...
Epoch: 67 Training Loss: 3.257892 Validation Loss: 2.795484 Training Accuracy: 0.957887 Validation Accuracy: 0.958755
Epoch: 68 Training Loss: 3.336628 Validation Loss: 3.072439 Training Accuracy: 0.955744 Validation Accuracy: 0.960659
Epoch: 69 Training Loss: 3.362346 Validation Loss: 2.784997 Training Accuracy: 0.958096 Validation Accuracy: 0.965714
Epoch: 70 Training Loss: 3.220004 Validation Loss: 3.289794 Training Accuracy: 0.961220 Validation Accuracy: 0.952088
Epoch: 71 Training Loss: 3.320972 Validation Loss: 3.198232 Training Accuracy: 0.957381 Validation Accuracy: 0.958901
Epoch: 72 Training Loss: 3.536178 Validation Loss: 3.194713 Training Accuracy: 0.957411 Validation Accuracy: 0.952088
Epoch: 73 Training Loss: 3.426154 Validation Loss: 3.120841 Training Accuracy: 0.958631 Validation Accuracy: 0.958901
Epoch: 74 Training Loss: 3.265437 Validation Loss: 3.356003 Training Accuracy: 0.957887 Validation Accuracy: 0.956996
Epoch: 75 Training Loss: 3.439150 Validation Loss: 3.091171 Training Accuracy: 0.957649 Validation Accuracy: 0.956996
Epoch: 76 Training Loss: 3.465368 Validation Loss: 2.794633 Training Accuracy: 0.954078 Validation Accuracy: 0.962857
Epoch: 77 Training Loss: 3.486829 Validation Loss: 3.017923 Training Accuracy: 0.951697 Validation Accuracy: 0.960806
Epoch: 78 Training Loss: 3.612007 Validation Loss: 3.214564 Training Accuracy: 0.954286 Validation Accuracy: 0.956996
Epoch: 79 Training Loss: 3.410268 Validation Loss: 3.353851 Training Accuracy: 0.954792 Validation Accuracy: 0.953846
Epoch: 80 Training Loss: 3.412967 Validation Loss: 3.363153 Training Accuracy: 0.957619 Validation Accuracy: 0.950329
Epoch: 81 Training Loss: 3.434409 Validation Loss: 3.114588 Training Accuracy: 0.957143 Validation Accuracy: 0.961758
Epoch: 82 Training Loss: 3.435157 Validation Loss: 2.891220 Training Accuracy: 0.960476 Validation Accuracy: 0.957143
Epoch: 83 Training Loss: 3.539886 Validation Loss: 3.095615 Training Accuracy: 0.951905 Validation Accuracy: 0.958755
Epoch: 84 Training Loss: 3.255761 Validation Loss: 2.973506 Training Accuracy: 0.960536 Validation Accuracy: 0.959048
Epoch: 85 Training Loss: 3.418598 Validation Loss: 3.146672 Training Accuracy: 0.955030 Validation Accuracy: 0.955092
Epoch: 86 Training Loss: 3.410447 Validation Loss: 2.851292 Training Accuracy: 0.955000 Validation Accuracy: 0.966520
Epoch: 87 Training Loss: 3.395330 Validation Loss: 3.016580 Training Accuracy: 0.955476 Validation Accuracy: 0.962711
Epoch: 88 Training Loss: 3.150888 Validation Loss: 3.139933 Training Accuracy: 0.960268 Validation Accuracy: 0.960000
Epoch: 89 Training Loss: 3.449000 Validation Loss: 3.116051 Training Accuracy: 0.958601 Validation Accuracy: 0.960952
Epoch: 90 Training Loss: 3.275479 Validation Loss: 3.254880 Training Accuracy: 0.958334 Validation Accuracy: 0.957949
Epoch: 91 Training Loss: 3.290689 Validation Loss: 2.742065 Training Accuracy: 0.955715 Validation Accuracy: 0.966667
Epoch: 92 Training Loss: 3.631882 Validation Loss: 2.903477 Training Accuracy: 0.952619 Validation Accuracy: 0.963663
Epoch: 93 Training Loss: 3.430934 Validation Loss: 3.000058 Training Accuracy: 0.958125 Validation Accuracy: 0.955604
Epoch: 94 Training Loss: 3.366206 Validation Loss: 3.032690 Training Accuracy: 0.957709 Validation Accuracy: 0.959048
Epoch: 95 Training Loss: 3.562633 Validation Loss: 3.032979 Training Accuracy: 0.952857 Validation Accuracy: 0.960659
Epoch: 96 Training Loss: 3.502647 Validation Loss: 3.049100 Training Accuracy: 0.953810 Validation Accuracy: 0.956703
Epoch: 97 Training Loss: 3.353789 Validation Loss: 2.807156 Training Accuracy: 0.960953 Validation Accuracy: 0.961465
Epoch: 98 Training Loss: 3.397426 Validation Loss: 3.192062 Training Accuracy: 0.955239 Validation Accuracy: 0.959853
Epoch: 99 Training Loss: 3.418821 Validation Loss: 2.892588 Training Accuracy: 0.954762 Validation Accuracy: 0.965421
Epoch: 100 Training Loss: 3.282185 Validation Loss: 3.099401 Training Accuracy: 0.957887 Validation Accuracy: 0.955897
plt.figure(figsize=(10,8))
plt.plot(train_losses,'-',linewidth = 2.5, label='Training loss')
plt.plot(valid_losses,'-', linewidth = 2.5, label='Validation loss')
plt.xlabel('Epochs',fontsize = 20)
plt.legend(fontsize = 14)
plt.xticks(fontsize = 12)
plt.yticks(fontsize = 12);
plt.figure(figsize=(10,8))
plt.plot(train_acc,'-', linewidth = 2.5, label='Training accuracy')
plt.plot(valid_acc,'-', linewidth = 2.5, label='Validation accuracy')
plt.xlabel('Epochs', fontsize = 20)
plt.legend(fontsize = 14)
plt.xticks(fontsize = 12)
plt.yticks(fontsize = 12);
vgg19.load_state_dict(torch.load('./../models/vgg19_pneumonia.pt'))
<All keys matched successfully>
# track test loss
test_loss = 0.0
class_correct = list(0. for i in range(2))
class_total = list(0. for i in range(2))
vgg16.eval()
# iterating over test data
for data, target in tqdm(test_loader, desc = 'Model testing: '):
# moving tensors to GPU if CUDA is available
if train_on_gpu:
data, target = data.cuda(), target.cuda()
# forward pass: computing predicted outputs by passing inputs to the model
output = vgg19(data)
target = target.long()
# calculating the batch loss
loss = criterion(output, target)
# updating test loss
test_loss += loss.item()*data.size(0)
# converting output probabilities to predicted class
_, pred = torch.max(output, 1)
# comparing predictions to true label
correct_tensor = pred.eq(target.data.view_as(pred))
correct = np.squeeze(correct_tensor.numpy()) if not train_on_gpu else np.squeeze(correct_tensor.cpu().numpy())
# calculating test accuracy for each object class
for i in range(len(target.data)):
label = target.data[i]
class_correct[label] += correct[i].item()
class_total[label] += 1
# average test loss
test_loss = test_loss/len(test_loader.dataset)
print('Test Loss: {:.6f}\n'.format(test_loss))
for i in range(2):
if class_total[i] > 0:
print('Test Accuracy of %5s: %2d%% (%2d/%2d)' % (
classes[i], 100 * class_correct[i] / class_total[i],
np.sum(class_correct[i]), np.sum(class_total[i])))
else:
print('Test Accuracy of %5s: N/A (no training examples)' % (classes[i]))
print('\nTest Accuracy (Overall): %2d%% (%2d/%2d)' % (
100. * np.sum(class_correct) / np.sum(class_total),
np.sum(class_correct), np.sum(class_total)))
Test Loss: 0.196179 Test Accuracy of PNEUMONIA: 84% (197/234) Test Accuracy of NORMAL: 96% (377/390) Test Accuracy (Overall): 91% (574/624)
confusion_matrix = torch.zeros(2,2)
with torch.no_grad():
for i, (data, target) in enumerate(test_loader):
if train_on_gpu:
data, target = data.cuda(), target.cuda()
output = vgg19(data)
target = target.long()
_, preds = torch.max(output, 1)
for t, p in zip(target.view(-1), preds.view(-1)):
confusion_matrix[t.long(), p.long()] += 1
print(confusion_matrix)
tensor([[197., 37.],
[ 13., 377.]])
cm = confusion_matrix.numpy()
tn, fp, fn, tp = cm.ravel()
precision = tp/(tp+fp)
recall = tp/(tp+fn)
f1_score = (2*precision*recall)/(precision+recall)
print("Precision: {:.4f}\nRecall: {:.4f}\nF1 Score: {:.4f}".format(precision,recall,f1_score))
Precision: 0.9106 Recall: 0.9667 F1 Score: 0.9378
plt.figure(figsize=(8,7))
xticks = np.array(['NORMAL','PNEUMONIA'])
sns.heatmap(cm, annot = True,
xticklabels = xticks, yticklabels = xticks)
plt.xlabel('PREDICTED',fontsize = 15)
plt.ylabel('TRUE',fontsize = 15);
# obtaining one batch of test images
data, target = next(iter(test_loader))
data.numpy()
# moving model inputs to cuda, if GPU available
if train_on_gpu:
data = data.cuda()
# getting sample outputs
output = vgg19(data)
# converting output probabilities to predicted class
_, preds_tensor = torch.max(output, 1)
preds = np.squeeze(preds_tensor.numpy()) if not train_on_gpu else np.squeeze(preds_tensor.cpu().numpy())
# plotting the images in the batch, along with predicted and true labels
fig = plt.figure(figsize=(20,10))
for idx in np.arange(8):
ax = fig.add_subplot(2,4, idx+1, xticks=[], yticks=[])
ax.imshow(data[idx].permute(1, 2, 0) if not train_on_gpu else data[idx].cpu().permute(1, 2, 0))
ax.set_title("{} ({})".format(classes[preds[idx]], classes[target[idx]]),
color=("green" if preds[idx]==target[idx] else "red"))
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers). Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
